AI Monthly digest #5 – AlphaStar beats human champions, robots learn to grasp and a Finnish way to make AI a commodity

AI Monthly digest #5 – AlphaStar beats human champions, robots learn to grasp and a Finnish way to make AI a commodity

/in , , /by and

With a ground-breaking AlphaStar performance, January kickstarted 2019 AI-related research and activities.

The January edition of AI Monthly digest brings a marvel to behold–AlphaStar beating human champions at the real-time strategy game StarCraft. To find out why that’s important, read the full story.

AlphaStar beats top world StarCraft II pro players

AlphaStar, a DeepMind-created agent, beat two world-famous professional players in StarCraft II. The agent was using a Protoss race and playing against another Protoss. The game itself can also play Zerg and Terrans, but AlphaStar is trained only to play Protoss vs. Protoss matches.
The machine defeated Dario “TLO” Wunsch, a Zerg specialist playing Protoss for the occasion, 5-0 in the first five-match round. It then made quick work of professional Protoss player Grzegorz “MaNa” Komnicz beating the champion 5:0.
A noticeable advantage AlphaStar had against both players was its access to the entire StarCraft II map at once. It is still obscured by the fog of war, but the agent doesn’t have to mimic the human’s camera moves. DeepMind prepared one other agent to address this issue and plays using a camera interface, but it lost to MaNa 0:1.
To make the matches fair, the DeepMind team reduced the Actions Per Minute (APM) ratio to a human level and ensured the machine had no advantage in terms of reaction time. Nonetheless, it was clear at crucial moments that AlphaStar had bursts of APM far above human abilities. DeepMind is aware of this and will probably do something about it in the future. For now, however, we will be content to focus on what we have seen.

How the matches went

Unlike human players, AlphaStar had employed some unorthodox yet not necessarily wrong strategies – avoiding walling the entrance to the base with buildings was the most conspicuous one. What’s more, the model used significantly more harvesting drones than pro players normally use.

Beyond its superiority in micromanagement (the art of managing a single unit and using its abilities on the battlefield), the agent didn’t display any clearly non-human strategies or abilities. However, AlphaStar was seen at its finest when it managed to win the match by managing a large number of Stalkers, the unit that is normally countered by Immortals in a rock-paper-scissors manner. As MaNa, the human player confronting the agent, noted, he had never encountered a player with such abilities. As such, the gameplay was clearly on a superhuman level, especially considering the fact that MaNa executed the counter-tactic, which failed due to AlphaStar’s superior micromanagement.

How the Deepmind team did it

The initial process of training the agent took ten days – three of supervised learning built on the basis of replays of top StarCraft II players. The team then infused the agent with reinforcement learning abilities (an approach similar to our team’s cracking Montezuma’s Revenge) and created the “AlphaStar” league to build multiple agents competing against each other. league witnessed a similar cycle with some strategies emerging and being later countered.

Related:  Five top artificial intelligence (AI) trends for 2019

After that, the team selected five agents for a match with TLO. To further polish their skills, the agents were trained for another week before the match with the MaNa. As a Protoss specialist, MaNa posed a greater challenge than TLO, a Zerg-oriented player who was learning Protoss tactics only to square off against AlphaStar.
Courtesy of Blizzard, the developer of StarCraft II, Deepmind was delivered a significantly faster version of StarCraft II. This version enabled each agent in AlphaStar league to experience up to 200 years of real-time gameplay in just two weeks.

Why it matters

The AI community has grown accustomed to witnessing agents cracking Atari Classics and popular board games like chess or Go. Both environments provide a diverse set of challenges, with chess being a long-term fully observable strategy game and Atari delivering real-time experience with limited data.
StarCraft combines all manner of challenge by forcing players to follow the long-term strategy without knowledge of an opponent’s strategy and movement until it is in the line of sight of its own units (normally the battlefield is covered by the “fog of war”). Each encounter may show that a strategy needs to be fixed or adapted, as many units and strategies tend to work in a rock-paper-scissors manner, enabling players to play in a tactic-counter-tactic circle. Problem-solving in real time while sticking to a long-term strategy, constantly adapting to a changing environment and optimizing one’s efforts are all skills that can be later extrapolated to solve more challenging real-world problems.

Related:  AI Monthly digest #3 - artificial intelligence in science getting big

Thus, while playing computer games is fun, the research they enable is very serious. It also lays bare the contrast between human and machine abilities. The computer was able to beat a human player after about four hundred years of constant playing. The human expert, meanwhile, was twenty-five years old, had started playing StarCraft at the age of six and had to sleep or go to school while not playing StarCraft.
Nevertheless, impressive and inspiring.

Understanding the biological brain using a digital one

According to Brain Injury Alliance Wisconsin, approximately 10% of individuals are touched by brain injuries and 5.3 million Americans (a little more than 2% of the US population) live with the effects of a brain injury. Every 23 seconds someone suffers a brain injury in the US.
Such injuries add up to $76.5 billion in annual costs once treatment, transportation and the range of indirect costs like lost productivity are considered.
While brain trauma may sometimes be responsible for the loss of speech, strokes and motor neurone disease are also to blame. Although patients lose the ability to communicate, they often remain conscious. Stephen Hawking is perhaps the most famous such person. Hawking used a speech generator, which he controlled with the muscles in his cheek. The generators can also be controlled with the eyes.
Applying neural networks to interpret the signals within the brain enabled the scientists to reconstruct speech. Summarizing the efforts, Science magazine points out that the effects are more than promising.
Alzheimer’s disease is another challenge that may be tackled with neural networks. There are no medications that heal the disease, but applying the treatment early enough makes it manageable. With Alzheimer’s, the earlier the diagnosis is made, the more effective the treatment will be. The challenge is in the diagnosis, which often comes too late for the disease to be reversible.

Related:  Powering up demand forecasting with machine learning

By feeding the neural networks with glucose PET scans, researchers from the University of California delivered a system that can diagnose the early symptoms of Alzheimer’s disease up to six years earlier than doctors do.

Why it matters

The human brain is one of the most complex devices in the universe, so understanding how it works is obviously a great challenge. Applying neural networks to treat brain-related diseases may come with a bit of irony – we need an outer, artificial brain to outthink the way our own is working.

Democratizing the AI – the Finnish way

Machine learning and artificial intelligence, in general, tend to be depicted as a black box, with no way to get to know “what the machine is thinking”. At the same time, it is often shown as a miraculous problem-solver, pulling working solutions out of seemingly nothing like a magician procuring a rabbit from a hat. But this too is a misconception.
Like every tool before it, neural networks need to be understood if they are to yield the most valuable outcomes. That’s one reason Finland aims to train its population in AI techniques and machine learning. Starting with 1% of its population (or roughly 55,000 people), the country aims to boost its society and economy by being a leader in the practical application of AI.
Initially a grassroots movement, the initiative gained the support of the government and Finland’s largest employers.

Why it matters

The biggest barrier in using AI and machine learning-powered techniques is uncertainty and doubt. Considering that people are afraid of things they don’t understand, spreading the knowledge about machine learning will support the adoption and reduce societal reluctance to adapting these tools. Moreover, understanding the mechanisms powering ML-based tools will give users a greater understanding of just what the tools are and are not capable of.

New state-of-the-art in robotic grasping

The issues Artificial Intelligence prompts frequently ignite philosophical debate and add interesting insight and inspiration. This recent paper on robot grasping is short of neither insights nor inspiration.
The idea behind the use of reinforcement learning to control robotic arms is simple - hard-coding all the possible situations the robot may encounter is virtually impossible, but building a policy to follow is much easier Click To Tweet
What’s more, building the controller for the robotic arm requires the mountains of data coming from the sensors to be cross-combined. Every change – be it lighting, color or position of an object — can confuse the controller and result in failure.
Thus, the research team built a neural network that processes the input into the “canonical” version, stripped of the insignificant details like shades or graphical patterns – so that grasping is the only thing that matters. Ushering in a new state of the art in robotic grasping, the results are impressive.

Why do the results matter?

There are two reasons these results are important. First, building the controllers of robotic arms is now simpler. Robots that can move in non-supervised, non-hardcoded ways and grasp objects will be used in astonishing ways to improve human lives–for example, as assistants for the disabled or by augmenting the human workforce in manufacturing.
The second breakthrough is how researchers achieved their improvements. Instead of building more powerful neural networks to improve the input processing, the researchers downgraded the data into a homogenous, simplified “canonical” version of the reality. It seems that when it comes to robotic perception, Immanuel Kant was right. There are “things that exist independently of the senses or perception”, but they are unknowable–at least for a robotic observer. Only operating within a simplified reality enables the robot to perform the task.

Keep informed on state-of-the-art machine learning

With the rapidly changing ML landscape, it is easy to lose track of the latest developments. A lecture given by MIT researcher Lex Fridman is a good way to start. The video can be seen here:

Read previous editions of AI Monthly digest:

How artificial intelligence can fight hate speech in social media

How artificial intelligence can fight hate speech in social media

/in , /by

With social media users numbering in the billions, all hailing from various backgrounds and bringing diverse moral codes to today’s wildly popular platforms, a space for hate speech has emerged. Internet service providers have responded by employing AI-powered solutions to address this insidious problem.

Hate speech is a serious issue. It undermines the principles of democratic society and the rules of public debate. Legal views on the matter vary. On the internet, every statement that transgresses the standards for hate speech established by a given portal (Facebook, Twitter, Wikipedia etc.) may be banned from publication. To get around such bans, numerous groups have launched platforms to exchange their thoughts and ideas. Stricter definitions of hate speech are common. They make users feel safe, which is paramount for social media sites as the presence of users is often crucial to income. And that’s where building machine learning models spotting the hate speech comes in.

What is hate speech?

The definition of hate speech varies by country and may apply to various aspects of language. Laws prohibit directing hateful speech and defamatory language toward another’s religion, ethnicity or sexuality. Many countries penalize anyone who agitates violence or genocide. Additionally, many legislatures ban symbols of totalitarian regimes and limit the freedom of assembly when ideologies like fascism or communism are involved.

Related:  Five top artificial intelligence (AI) trends for 2019

In its most common form, hate speech attacks a person or group based on race, religion, ethnicity, national origin, disability, gender or sexual orientation. As regards what’s legal, the devil, as usual, is in the details. Finding the balance between freedom of speech and the protection of minority rights makes it difficult to produce a strict definition of hate speech. However, the problem has certainly grown with the rise of social media companies. The 2.27 bln active users of Facebook, who come from various backgrounds and bring diverse moral codes to the platform, have unwittingly provided a space for hate speech to emerge. Due to the international and flexible nature of the Internet, battling online hate speech is a complex task involving various parties.
Finally, there is a proven link between offensive name-calling and higher rates of suicide within migrant groups.

Why online hate speech is a problem

As a study from Pew Research Center indicates, 41% of American adults have experienced some form of online harassment. The most common is offensive name calling (experienced by 27%) and purposeful embarrassment (22%). Moreover, a significant number of American adults have experienced physical threats, sustained harassment, stalking and sexual harassment (10%, 7%, 7% and 6% respectively).
How artificial intelligence can fight hate speech in social media
Hate speech itself has serious consequences for online behavior and general well-being. 95% of Americans consider it a serious problem. At 27%, more than one in four Americans have reported deciding not to post something when encountering hate speech toward another user. 13%, meanwhile, have stopped using a certain online platform after witnessing harassment. Ironically, protected as a form of free speech, hate speech has resulted in muting more than a quarter of internet users.

Who should address the issue

Considering both vox populi and practice, online platforms are to tackle the problem of user’s hate speech. According to the Pew Research Center report cited above, 79% of Americans say that online service and social network providers are responsible for addressing harassment. In Germany, companies may face a fine of up to 50m euro if they fail to remove within 24 hours illegal material, including fake news and hate speech.

Related:  AI Monthly digest #4 - artificial intelligence and music, a new GAN standard and fighting depression

Hate speech is not always as blatant as calling people names. It can come in many subtler forms, posing as neutral statements or even care. That’s why building more sophisticated AI models that can recognize even the subtlest forms of hate speech is called for.

How those models should be built

When building a machine learning-powered hate speech detector, the first challenge is to build and label the dataset. Given that the differences between hate speech and non-hate speech are highly contextual, constructing the definition and managing the dataset is a huge challenge. The context may depend on:

  • The context of the discussion –  historical texts full of outdated expressions may be automatically (yet falsely) classified as hate speech
    • Example: Mark Twain’s novels use insulting language; citing them may set off  hate speech bells.
  • How the language is used – in many countries, hate speech used for artistic purposes is tolerated.
    • Example: Hip-hop often uses misogynistic language while heavy metal (especially the more extreme sub-genres) is rife with anti-religious lyrics.
  • The relationship of the speaker to the group being hated – the members of a group are afforded more liberties with using aggressive or insulting language when addressing other members of that group than are those who are not a part of it.
    • Example: the term “sans-cullottes” was originally coined to ridicule the opponents of conservatives. It literally meant “people with no trousers” and was aimed at the working class, members of whom wore long trousers instead of the fashionable short variety. The term went on to enter the vernacular of the working classes in spite of its insulting origins.

Irony and sarcasm pose yet another challenge. According to Poe’s law, without smileys or other overt signs from the writer, ironic statements made online are indistinguishable from serious ones. In fact, the now-ubiquitous emoticons were invented by professors at Carnegie Mellon University to avoid mistakes.
When the dataset is ready, building a blacklist of the most common epithets and hate-related slurs may be helpful, as automation-based blacklist models are effective 60% of the time in spotting online hate speech (based on our in-house benchmarks). Building both supervised and unsupervised learning models to spot the new combinations of harmful words or finding existing ones may raise that effectiveness further. Hate speech is dynamic and thus evolves rapidly as new forms and insulting words emerge. By keeping an eye on conversations and general discourse, machine learning models can spot suspicious new phrases and alert administration.

A formula of hate

An automated machine learning model is able to spot the patterns of hate speech based on word vectors and the positions of words with certain connotations. Thus, it is easier to spot emerging hate speech that went undetected earlier, as current politics or social events may trigger new forms of online aggression.
Unfortunately, people spreading hate have shown serious determination to overcome automated systems of spotting hate speech by combining common ways of fooling machines (like using acronyms and euphemisms) and perpetuate hate.

Challenges and concerns

One of the main concerns in building machine learning models is finding a balance between the model’s vigilance and the number of false positives it returns. Considering the uneasy relations between hate speech and freedom of speech, producing too many false positives may be considered by users an attempt at censorship.
Another challenge is to build the dataset and label the data to train the model to recognize hate speech. As machines themselves are truly neutral, the person responsible for the dataset may be biased or at least influenced to profile the hate speech recognition model. Thus, the model may be built to purposefully produce false-positives in order to reduce the prevalence of certain views in a discussion.

Five top artificial intelligence (AI) trends for 2019

Five top artificial intelligence (AI) trends for 2019

/in , /by

As the recently launched AI Monthly digest shows, significant improvements, breakthroughs and game-changers in machine learning and AI are months or even weeks away, not years. It is, therefore, worth the challenge to summarize and show the most significant AI trends that are likely to unfold in 2019, as machine learning technology becomes one of the most prominent driving forces in both business and society.
According to a recent Deloitte study, 82% of companies that have already invested in AI have gained a financial return on their investment. For companies among all industries, the median return on investment from cognitive technologies is 17%.
AI is transforming daily life and business operations in a way seen during previous industrial revolutions. Current products are being enhanced (according to 44% of respondents), internal (42%) and external (31%) operations are being optimized and better business decisions are being made (35%).
With that in mind, it is better to see the “Trend” as a larger and more significant development than a particular technology or advancement. That’s why chatbots or autonomous cars are not so much seen as particular trends, but rather as separate threads in the fabric that is AI.
That distinction aside, here are five of the most significant and inspiring artificial intelligence trends to watch in 2019.

1. Chatbots and virtual assistants ride the lightning

AI Trends 2019 - chatbot
The ability to process natural language is widely considered a hallmark of intelligence. In 1950, Alan Turing proposed his famous test to determine if a particular computer is intelligent by asking the ordinary user to determine if his conversational partner is a human or a machine.
The famous test was initially passed in 1966 by ELIZA software, though it had nothing to do with natural language processing (NLP) – it was just a smart script that seemed to understand text. Today’s NLP and speech recognition solutions are polished enough not only to simulate understanding but also to produce usable information and deliver business value.
While still far from perfect, NLP has gained a reputation among businesses embracing chatbots. PwC states that customers prefer to talk with companies face-to-face but chatbots are their second preferred channel, slightly outperforming email. With their 24/7 availability, chatbots are perfect for emergency response (46% of responses in the PwC case study), forwarding conversations to the proper employee (40%) and placing simple orders (33%). Juniper Research predicts that chatbots will save companies up to $8bln annually by 2022.
NLP is also used in another hot tech trend–virtual assistants. According to Deloitte, 64% of smartphone owners say they use their virtual assistant (Apple Siri, Google’s Assistant) compared to 53% in 2017.
Finally, Gartner has found that up to 25% of companies will have integrated a virtual customer assistant or a chatbot into their customer service by 2020. That’s up from less than 2% in 2017.

2. Reducing the time needed for training

AI Trends 2019 training time reduction
Academic work on AI often focuses on reducing the time and computing power required to train a model effectively, with the goal of making the technology more affordable and usable in daily work. The technology of artificial neural networks has been around for a while (theoretical models were designed in 1943), but it works only when there are enough cores to compute machine learning models. One way to ensure such cores are present is to design more powerful hardware, though this comes with limitations. Another approach is to design new models and improve existing ones to be less computing hungry.
AlphaGo, the neural network that vanquished human GO champion Lee Sidol, required 176 GPUs to be trained. AlphaZero, the next iteration of the neural network GO phenom, gained skills that had it outperforming AlphaGo in just three days using 4 TPUs.
Expert augmented learning is one of most interesting ways to reduce the effort required to build reinforcement-based models or at least ones that are reinforcement learning-enhanced. Contrary to policy-blending, expert augmented learning allows data scientists to channel their knowledge not only from another neural network but also from a human expert or another machine. Researchers at deepsense.ai have recently published a paper on using transfer learning to break Montezuma’s Revenge, a game that reinforcement learning agents had long struggled to break.
Another way to reduce the time needed to train a model is to optimize the hardware infrastructure required. Google Cloud Platform has offered a cloud-based tailored environment for building machine learning models without the need for investing in on-prem infrastructure. Graphics card manufacturer Nvidia is also pushing the boundaries, as GPUs tend to be far more effective in machine learning than CPUs.
Yet another route is to scale and redesign the architecture of neural networks to use existing resources in the most effective way possible. With its recently developed GPipe infrastructure, Google has been able to significantly boost the performance of Generative Adversarial Networks on an existing infrastructure. By using GPipe, researchers were able to improve the performance of ImageNet Top-1 Accuracy (84.3% vs 83.5%) and Top-5 Accuracy (97.0% vs 96.5%), making the solution the new state-of-the-art.

3. Autonomous vehicles’ speed rising

AI trends 2019 autonomous cars
According to PwC estimates, 40% of mileage in Europe could be covered by autonomous vehicles by 2030. Currently, most companies are still developing the technology behind these machines. We are proud to say that deepsense.ai is contributing to the push. The process is driven mostly by the big social and economic benefits involved in automating as many driving processes as possible.
According to the US Department of Transportation, 63.3% of the $1,139 billion of goods shipped in 2017 were moved on roads. Had autonomous vehicles been enlisted to do the hauling, the transport could have been organized more efficiently, and the need for human effort vastly diminished. Machines can drive for hours without losing concentration. Road freight is globally the largest producer of emissions and consumes more than 70% of all energy used for freight. Every optimization made to fuel usage and routes will improve both energy and time management.
The good news here is that there are already advanced tests of the technology. Volvo has recently introduced Vera, the driverless track aimed at short-haul transportation in logistics centers and ports. Its fleet of cars is able to provide a constant logistics stream of goods with little human involvement.
In a related bid, US grocery giant Kroger recently started tests of unmanned delivery cabs, sans steering wheel and seats, for daily shopping. Bolder still are those companies (including Uber) testing their autonomous vehicles on the roads of real towns, while others build models running in sophisticated simulators.
With Kroger, Uber and Google leading the way, other companies are sure to fall into line behind them, forming one of most important AI trends 2019.

4. Machine learning and artificial intelligence will be democratized and productionized

AI trends 2019 machine learning productization
There would be no machine learning without data scientists, of which there remain precious few, at least of the skilled variety. Job postings for data scientists rose 75% between 2015 and 2018 at indeed.com while job searches for this position rose 65%. According to Glassdoor data, data scientist was the hottest job in 2018. Due to the popularization of big data, artificial intelligence and machine learning, the demand for data science professionals will continue to rise. And that not only enterprise but also scientific researchers seek their skills certainly bodes well for the profession.
Despite being associated with high-tech companies, machine learning techniques are becoming more common in solving science-related problems. In the last quarter of 2018, Deepmind unveiled a tool to predict the way proteins fold. Another project enabled scientists to derive the laws of physics from fictional universes.
According to O’Reilly data, 51% of surveyed organizations already use data science teams to develop AI solutions for internal purposes. The adoption of AI tools will no doubt be one of the most important AI trends in 2019, especially as business and tech giants are not the only organizations using AI in their daily work.

5. AI responsibility and transparency

AI trends 2019 models transparency
Last but not least, as the impact of machine learning on business grows, so too does the social and legal impact. On the heels of the first fatal accident involving an autonomous car, the question of who is responsible for crashes and the famous trolley problem are getting more important.
At issue here, first and foremost, is hidden bias in data sets, a problem for any company using AI to power-up daily operations. That includes Amazon, which had put AI in charge of preprocessing resumes. Trained with 10 years worth of various resumes, the system was unintentionally biased against women applying for tech positions. With the rising adoption of machine learning models in various industries, the transparency of artificial intelligence will be on the rise. The issue of countering bias unconsciously developed within datasets and taken by machine learning models as truth incarnate is being discussed seriously by tech giants like Salesforce. The machine learning community has also taken up the problem: there is a Kaggle competition aimed at building unbiased and cultural context-agnostic image recognition models to use in computer vision.
Finally, as I alluded earlier, the question of who is responsible for actions taken by AI-powered devices and the famous trolley problem are both moving to the fore. If a self-driving car had a choice, should it hit an elderly person or a child? Focus on saving the life of a driver or a person walking by? According to a global study, the answers depend heavily on the culture the responder grew up in. When facing the extreme situation of a car accident today, it is the driver who is solely responsible for his or her choices. When the car is autonomous, however, and controlled by a virtual agent, all the choices are made by a neural network, which raises some very unsettling questions.
Of course, such problems are not confined to the realm of autonomous vehicles. Machine learning-powered applications are getting more and more attention as a tool for supporting medical treatment. Medical data is expected to rise at an exponential rate, with a compound annual growth rate of 36%. Considering the high level of standardization within diagnostic data, medical data is ripe for utilizing machine learning models, which can be employed to augment and support the treatment process.
When thinking about AI trends 2019, bank on more transparent and socially responsible models being built.

The take-away – the social context will be central to AI trends 2019

No longer are AI and machine learning confined to pure tech; they now have an impact on entire businesses and the whole of society. The common comparison with the steam engine revolution is an apt one – machine learning models will digitally transform both big and small business in ways never before seen.
AI trends 2019 infographic
Given that, picking AI trends 2019 only by selecting technology would be to miss a vital aspect of ongoing changes. That is, they are as ubiquitous as they are far-reaching.

AI Monthly digest #4 - artificial intelligence and music, a new GAN standard and fighting depression

AI Monthly digest #4 – artificial intelligence and music, a new GAN standard and fighting depression

/in , , /by and

December brought both year-end summaries and impressive research work pushing the boundaries of AI.

Contrary to many hi-tech trends, artificial intelligence is a popular term in society and there are many (not always accurate) views on the topic. The December edition of AI Monthly digest both shares the news about the latest developments and addresses doubts about this area of technology.
Previous editions of AI Monthly digest:

Tuning up artificial intelligence and music

Machines are getting better at image recognition and natural language processing. But several fields remain unpolished, leaving considerable room for improvement. One of these fields is music theory and analytics.
Music has different timescales, as some parts repeat at scale of seconds while others extend throughout an entire composition and sometimes beyond. Moreover, music composition employs a great deal of repetition.

Google’s Magenta-designed model leverages the relative attention to spot how far two tokens (motifs) are, and produced convincing and quite relaxing pieces of piano music. The music it generated generally evokes Bach more than Bowie, though.
Researchers provided samples of both great and flawed performances. Although AI-composed samples still refer to classical music, there are more jazz-styled improvisations.

Related:  Don’t waste the power. AI supply chain management

Further information about studies on artificial music can be found on Magenta’s Music Transformer website.
Why does it matter?
Music and sound is another type of data machine learning can analyze. Pushing research on music further will deepen our knowledge of music and styles in a way that AI made Kurt Vonnegut’s dream of analyzing literature a reality.
Furthermore, the music industry may be the next to leverage data and the knowledge and talents of computer scientists. Apart from tuning the recommendation engines for streaming services, they may contribute more to the creation of music. A keyboard is after all a musical instrument.

2. The machine style manipulator

Generating fake images from real ones is a thing. Generative Adversarial Networks enhance their training abilities by analyzing real images, generating fake ones and then training to be as good as possible in determining which is real and what is shown in the images.

The challenge neural networks and those who design them fact is in producing convincing images of people, cars, or anything else that is to be recognized by the networks. In a recent research paper, the group behind the “one hour of fake celebrity faces” project introduced a new neural network architecture that separates high-level attributes and stochastic variations. In the case of human images, the high-level attribute may be a pose while freckles or the hairdo are stochastic variations.
In a recent video, researchers show the results of applying a style-based generator and manipulating styles later to produce different types of images.

The results are impressive – researchers were able to produce convincing images of people from various ethnic backgrounds. By controlling different levels of styles, researchers were able to tune up everything on image – from gender and ethnicity to the shape of glasses worn.
Why does it matter?
That’s basically the next state-of-the-art in GAN networks, the best-performing image recognition technology used. Producing fancy-looking fake images of faces and houses can significantly improve the performance of image recognition models. Ultimately, however, this technology may be a life saver, especially when applied in medical diagnosis, for example in diabetic retinopathy.

3. AI failures – errare (not only) humanum est

2018 saw significant improvements in machine learning techniques and artificial intelligence proved even further how useful it is. However, using it in day-to-day human life will not be without its challenges.

In his 1896 novel “An Outcast of the Islands”, Joseph Conrad wrote “It’s only those who do nothing that make no mistakes”. This idea can also be applied to theoretically mistake-proof machines. Apart from inspiring successes, 2018 also witnessed some significant machine learning failures:

  • Amazon’s gender-biased AI recruiter – the machine learning model designed to pre-process the resumes sent to the tech giant overlooked female engineers due to bias in the dataset. The reason was obvious – the tech industry is male-dominated. As algorithms have neither common sense nor social skills, it assumed that women are just not a good match for the tech positions the company was trying to fill. Amazon ditched the flawed recruiting tool, yet the questions about hidden bias in datasets remain.
  • Uber’s fatal autonomous car crash – the story of fatal crash is a bitter lesson for all autonomous car manufacturers. Uber’s system not only detected the pedestrian it hit while driving, but also autonomously decided to proceed and ignore warnings, killing 49-year old Elaine Herzberg.
  • World Cup predictions gone wrong – The World Cup gave us another bitter lesson, this time for predictive analytics. While the model built to predict brackets may have been sophisticated, it failed entirely. According to its predictions, Germany should have met Brazil in the finals. Instead, the German team didn’t manage to get out of its group while Brazil bent the knee before South Korea. The final came down to France versus Croatia, an unthinkable combination, both for machine learning and football enthusiasts around the world. The case was further described in our blogpost about failure in predictive analytics.

More examples of AI failures can be found in the Synced Review Medium blogpost.
Why does it matter?
Nobody’s perfect. Including machines. That’s why users and designers need to be conscious of the need to make machine learning models transparent. What’s more, it is the next voice to ensure that machine learning model results are validated – a step that is tempting to overlook for early adopters.

4. Smartphone-embedded AI may detect the first signs of depression

A group of researchers from Stanford University has trained a model with pictures and videos of people who are depressed and people who are not. The model analyzed all the signals the subjects sent, including tone of voice, facial expressions and general behaviour. These were observed during interviews conducted by an avatar controlled by a real physician. The model proved effective in detecting depression more than 80% of the time. The machine was able to recognize slight differences between people suffering from depression and people who were not.
Why does it matter?
According to WHO, depression is the leading cause of disability worldwide. If not cured, it can lead to suicide, the second most common cause of death among 15-29-year-olds.

Related:  Powering up demand forecasting with machine learning

One barrier to helping people suffering from depression is inaccurate assessment. There are regions in the world where less than 10% of people have access to proper treatment. What’s more, mental illness is often stigmatized, and treatment is both costly and hard to access. These factors, and the fact that early symptoms are easily overlooked, lead many patients to avoid looking for care and medical support.
The experiment is a step toward building an automated and affordable system for spotting signs of depression early on, when the chance for a cure is highest.

5. So just how can AI hurt us?

Machine learning is one of the most exciting technologies of the twenty-first century. But science fiction and common belief have provided no lack of doomsday scenarios of AI harming people or even taking over the world. Dispelling the myths and disinformation and providing knowledge should be a mission all AI-developing companies. If you’re new to the discussion, here’s an essay addressing the threat of AI.
Why does it matter?
Leaving the doubts unaddressed may result in bias and prejudice when making decisions, both business and private ones. The key to making the right decisions is to be informed on all aspects of the issue. Pretending that Stephen Hawking’s and Elon Musk’s warnings about the cataclysmic risks AI poses were pointless would indeed be unwise.
On the other hand, the essay addresses less radical fears about AI, like hidden bias in datasets leading to machine-powered discrimination or allowing AI to go unregulated.
That’s why the focus on machine morality and the transparency of machine learning models is so important and comes up so frequently in AI Monthly digest.

Summary

December is the time to go over the successes and failures of the past year, a fact that applies equally to the machine learning community. Facing both the failures and challenges provides an opportunity to address common issues and make the upcoming work more future-proof.

Powering up demand forecasting with machine learning

Powering up demand forecasting with machine learning

/in /by

Demand forecasting is a field of predictive analytics, that aims to predict the demand of customers. It is done by analyzing statistical data and looking for patterns and correlations. Machine learning takes the practice to a higher level.
According to McKinsey study, AI-enhanced supply chain management may lead to improved accuracy by reducing forecasting errors by 20-50%.

Why demand forecasting is essential

There is always a context surrounding customer behavior. It may be an upcoming holiday, the weather or a recent event. As real product demand varies, businesses may face two challenges:

  • Income and profit loss when a product is out of stock or a service is unavailable
  • Cash tied up in stock or
  • The reduced margins that come with getting it out of warehouses

According to the Harvard Business Review, history is littered with missed demand forecasts that came with great losses for the business. To name a few of the most important:

  • In 1974, the U.S. energy sector made plans to double power generation due to a forecast calling for 7% annual growth in demand. But during the 1975-1985 period, growth faltered, coming in at a mere 2%. While many projects were canceled, the industry was heavily hit and consumer prices rose.
  • In 1983 and 1984, the PC market was forecast to explode and see 27-28 million units installed by 1988. As the story goes, by 1986 only 15 million units had been shipped and many PC manufacturers had gone out of business.

Considering the essential role of accurate demand forecasts, it is no surprise that an ARC global market study cited by Forbes predicts that demand applications will account for just under a third of the $2 billion supply planning suites market.

Human-powered demand forecasting methods

Demand is predicted using both qualitative and quantitative methods. In the most basic form, it is done by leveraging the experience of the seller or collective brainstorming in the company. Another popular way is to ask the customers. Conducting market surveys is one of the simplest methods of demand forecasting and may be applied in all industries, no matter the size or market segment. Other common methods include delphi forecasting method that relies on a panel of experts answering questions or employing game theory an leveraging.

Related:  Driverless car or autonomous driving? Tackling the challenges of autonomous vehicles

As companies are relying more and more on the data they collect, it is no surprise that traditional, human-based methods are being replaced by data-powered, automated systems. Using big data demand prediction is enabling a wide range of companies to leverage machine learning models in data exploration and extrapolation. If there is enough data to train a model, it is almost certain to outperform human data analysts and researchers. For proof of that, have a look at Google’s Sunroof project, which makes more accurate predictions of a roof’s sun exposure than any human analyst could do.

Demand prediction done right – case studies

Modern demand forecasting techniques go far beyond simple historical data analysis. When Netflix’s launched the design process for its series House of Cards, the company analyzed the most replayed parts of shows and films featuring Kevin Spacey, the show’s main star.

Powering up demand forecasting with machine learning - Capitol

When Netflix’s launched the design process for its series House of Cards, the company analyzed the most replayed parts of shows and films featuring Kevin Spacey, the show’s main star.

This helped it analyze what the target group was looking for more of. The data mining proved invaluable as the show’s second episode peaked at more than 4 million viewers. And according to the “Esquire”, the absence of Kevin Spacey in the last season “haunts the production” and the show is not the same anymore without the key actor.

Good demand forecasting can be a literal life-saver. In the UK, patients rely on the National Health Service’s (NHS) Blood & Transplant department, which consists of five manufacturing sites and 15 stock holding units. The department is responsible for transporting blood between hospitals, and managing both the donations and deliveries. The NHS’ previous system was relied heavily on manual data management, but it didn’t have the technology to analyze the data. Introducing a data mining-based system enabled the NHS to forecast demand by blood type and quantities it needed. With its new system, the NHS is able to prevent shortages by refreshing stock level readings every 30 minutes and connecting the information from various sources. Participating hospitals reported a 30% reduction in overstock waste and a 20% cut to delivery costs.

Summary

As demand forecasting is statistics-heavy and data-rich, it is ripe for the application of machine learning algorithms. The automation of data flows helps to manage logistics and find new ways to optimize an organization’s performance.

AI Monthly digest #3 - artificial intelligence in science getting big

AI Monthly digest #3 – artificial intelligence in science getting big

/in , /by and

November brought a lot of significant AI-related breaking news. Machine learning and deep learning models were folding proteins, deriving the laws of physics from fictional universes and mimicking the human brain in a way never before seen.

This edition of AI Monthly digest looks at scientific improvements made by AI with significant support from tech giants. Stories from November show that AI is clearly a great tool to improve countless lives – and the protein-folding model is the best example of solving problems thought to be unsolvable. Or at least since our universe came into existence.

1. Deepmind’s AlphaFold wins Protein-Folding Contest

It’s a bit ironic that the protein-based human brain designed silicon-based tools to better predict protein folding. That’s precisely the case with AlphaFold, the neural network designed by Deepmind to predict the shape of a protein from the DNA code.
The basic building blocks of all known life forms, proteins come in many forms, shapes and levels of complexity. Their shape and function are encoded within DNA. Yet the DNA itself contains only a sequence of amino acids that form the protein, and the way they will form a structure is not encoded in any way. Thus it is challenging to establish what the protein will look like. As Levintha’s Paradox states, it would take longer than the age of the universe to enumerate the possible configurations of a typical protein before reaching the right structure. Yet, the shape of protein is crucial in the treatment and diagnosis of numerous diseases, including Alzheimer’s, Parkinson’s and Hutchington’s.
Thanks to the relatively high amount of gene-related data, Deepmind was able to build a neural network that could predict the shape of proteins only from a DNA sequence. The algorithm analyzes the distance between particular amino acids and compares that with existing models, preparing estimations about possible shapes and folds.
The effects are seen as one of the most significant breakthroughs and an unprecedented progress on protein folding.
To read more about the model and how it works, see recent Deepmind’s blogpost.

2. Towards building an AI Physicist

A Physicist’s job is to create models of the universe that give us a better understanding of the reality surrounding us. Newton, Galileo, Archimedes and many others had to convert measurements and observations into the fundamental laws of the universe. The ability to discover important answers from data is the best proof of their genius.
Last month, MIT’s Tailin Wu and Max Tegmark presented work reporting on an “AI Physicist” deriving the laws of physics in artificial worlds. To crack the mysterious environments, the AI agent uses four strategies embedded in its architecture: divide-and-conquer, Occam’s razor, Unification and Lifelong Learning, which together allow it to discover and manipulate theories without supervision.
The system was able to produce correct theories for various simulated worlds created only to test its accuracy. The success is significant from a scientific point of view, as it proves that neural networks may be the tools to speed up physical science in more ways than we ever expected.
The Arxiv paper can be read here.

3. SpiNNaker – Human brain supercomputer runs for the first time

Building neural networks is about mimicking the human brain’s activity and processing abilities. Most common neural networks work by communicating in a constant way and exchanging information in a stream. The spiking neural network is more “biological” insofar as it works by forcing artificial neurons to communicate in spikes and exchange information in a “flash”. Instead of sending the information from point A to point B, it supports sending multiple bits of parallel information.

AI Monthly digest #3 - Artificial Intelligence in science getting big - Brain

The spiking neural network is more “biological” insofar as it works by forcing artificial neurons to communicate in spikes and exchange information in a “flash”

This system is mimicked in a SpiNNaker (Spiking Neural Network Machine) built at the University of Manchester’s School of Computer Science, backed by EPSRC and supported by the European Human Brain Project. The most significant outcome of the SpiNNaker is that it builds a working, small-scale model of the human brain, with the manifold scientific possibilities it brings with it.
More information may be read here.

4. New ImageNet state-of-the-art with GPipe

Since its foundations, the greatest challenge in computer science has been insufficient computing power. Multiplying the number of cores is one way to solve the problem, while optimizing the software is another. The challenge comes to bear in neural networks, as training a new one requires gargantuan computing power and no less time.
Thus, Google Brain’s GPipe is a significant improvement that makes neural networks more cost-effective. By using GPipe, neural networks can process significantly more parameters. And that leads to better results in training.
GPipe combines data parallelism and model parallelism, with a high level of automation and memory optimization. In the paper, researcher expanded the AmoebaNet from 155.3 million parameters to 557 million parameters and inserted as input 480×480 ImageNet Images. The result was an improvement in ImageNet Top-1 Accuracy (84.3% vs 83.5%) and Top-5 Accuracy (97.0% vs 96.5%), making the solution the new state-of-the-art.

5. Enterprise machine learning done right – Uber shows its best practices

With machine learning being a recent development, there are no proven and tested methodologies to build new AI-based developments nor best practices with which to dive in and use in development.
Uber’s recent blogpost shared an interesting vision of incorporating AI-driven culture into the company. Instead of building one large ML project to perform one task on an enterprise scale, Uber powers its teams with data scientists and looks for ways to automate their daily work. The company has now done a dozen projects, carried out by a variety of teams – from Uber Eats menu items ranking or marketplace forecasting to customer support.
On the back of this strategy, Uber has gone from a company that did not use machine learning at all to one now heavily infused with AI-based techniques. More details about ML applications in Uber can be seen in their blogpost.

Summary

AI, machine learning and deep learning tend to be seen as buzzwords, applied only by tech giants to solve their hermetic problems. Folding proteins, deriving the laws of physics and simulating the brain as it really works all show that AI is the rule-breaker and disruptor it any field it is applied in. Even if it was dominated by bold minds physicists of all time.

Driverless car or autonomous driving? Tackling the challenges of autonomous vehicles

Driverless car or autonomous driving? Tackling the challenges of autonomous vehicles

/in , /by

Among both traditional carmakers and cutting-edge tech behemoths, there is massive competition to bring autonomous vehicles to market.

It was a beautiful, sunny day of June 18, 1914 when the brilliant engineer Lawrence Sperry stunned the jury of Concours de la Securité en Aéroplane (Airline Safety Competition) by flying in front of their lodge with his hands held high. It was the first time the public had ever seen a gyroscopic stabilizer, one of the first autopiloting devices. Over a hundred years later, automatic flight control devices and maritime autopilots are common, while cars still require human operation. Thanks to machine learning and autonomous cars, that’s about to change.

What is the future of autonomous vehicles?

According to recent reports, autonomous cars are going to disrupt the private, public and freight transportation industries. A recent Deloitte publication reports that society is putting more and more trust in autonomous vehicles. In 2017, 74% of US, 72% of German and 69% of Canadian respondents declared that fully autonomous cars would not be safe. But those rates have now dropped significantly, to 47%, 45% and 44%, respectively.
Plans for building self-driving cars have been revealed by BMW, Nissan and Ford, while Uber and the Google-affiliated Waymo are also in the thick of the race. Companies aim both to build urban driving vehicles and autonomous trucks, while a startup scene supporting autonomous technology is emerging.
Thanks to the increasing popularity of autonomous cars, up to 40% of mileage could be driven in self-driving vehicles in 2030. But, as always, the devil is in the details.

What is an autonomous car?

To answer that question, the National Highway Traffic Safety Administration uses the autonomous vehicle taxonomy designed by the Society of Automotive Engineers, which lists five levels of automation.

  1. No automation – the driver performs all driving tasks
  2. Driver assistance – the car has built-in functions to assist the driver, who nonetheless must remain engaged in the driving process. Cruise control is one of the best examples.
  3. Partial automation – the vehicle has combined automated functions like acceleration and steering, but the driver must remain engaged. The gyroscopic stabilizer is an example of partial automation.
  4. Conditional automation – a human driver is necessary in totally unpredictable situations, but not required to monitor the environment all the time. BMW currently has a fleet of about 40 level 4 cars unleashed on testing grounds near Munich and in California.
  5. High automation – the car on this level may not even have a steering wheel and can deal with any situation encountered. Fully autonomous vehicles, which do not yet exist, occupy Level 5.

Building level 4 and 5 driverless vehicles is a great challenge because the driving process has a number of complicating factors. Unlike with a plane or ship, drivers usually have little to no time to respond to the changing environment. They must monitor the state of the machine, their surroundings, and the other drivers on the road. What’s more, any mistake can cause an accident – 37.133 people were killed in traffic accidents on American roads in 2017.
While we may not give ourselves credit as drivers, humans’ ability to process signals from various senses to control a car is a super power. It is not only about simply looking at the road – many drivers estimate the distance between cars by looking at reflections in the body of the car in front of it. Many drivers can hear changes in their engine’s performance or sense changing grip strength on various types of road.
To effectively replace human perception, sophisticated assistance systems rely on numerous sensors. GM’s report on autonomous cars  and driving technology safety lists:

  • Cameras – detect and track pedestrians and cyclists, monitor free space and traffic lights
  • Articulating radars –  detect moving vehicles at long range over a wide field of view
  • Short-range radars – monitor objects around the vehicle
  • Long-range radars – detect vehicles and measure velocity
  • Lidars – detect fixed and moving with objects high-precision laser sensors

Handling data from various sources that need to be processed in real time is a perfect task for deep neural networks, especially when it involves simultaneous work on non homogenous data taken from radars, images from cameras and lidar readings.
But building a system that automates driving is an enormous challenge, especially given the sheer number of serious decisions to be made when driving and the fact that a single bad decision can result in disaster.

Two ways how autonomous cars work

There are currently two approaches to building the models that control autonomous vehicles.
A component-based system – the controller is built with several independent models and software components each designed to handle one task, be it road sign recognition, managing the state of the vehicle or interpreting the sensors’ signals.

  • Pros –  dividing the system into subsystems makes building the software easier. Each component can be optimized and developed individually thus improving the system as a whole.
  • Cons – developing the model requires a massive amount of data to be gathered and processed. The image recognition module needs to be fed different data than the engine control device. This makes preparing the dataset to train more than a little challenging.  What’s more, the process of integrating the subsystems may be a challenge in and of itself.

End-to-end system – with this approach, a single model capable of conducting the entire driving process is built – from gathering information from the sensors to steering and reacting accordingly. deepsense.ai is moving ahead with just such a model.

  • Pros – it is easier to perform all the training within the simulation environment. Modern simulators provide the model with a high-quality, diverse urban environment. Using the simulated environment greatly reduces the cost of gathering data.

Although it is possible to label and prepare data gathered with the simulator, the technique requires a bit more effort. What’s more, it is possible to use a pre-trained neural network to mimic the simulated environment (a matrix of sorts) to further reduce the data to gather or generate. We expect the model to perform better than a component-based system would.

  • Cons – this type of model may be harder to interpret or reverse-engineer. When it comes to further tuning the model or reducing the challenge posed by the reality gap (see below) it may be a significant obstacle.

Facing the reality gap

Using a simulator-trained model in a real car is always challenging due to what is known as the reality gap. The reality gap represents all the differences and unexpected situations the model may encounter than the designer was able to predict and therefore prepare it for.
There are countless examples. The position of cameras in a real car may be different than in a simulated one, the simulation physics are necessarily incomplete, and there may be a hidden bug the model could exploit. Furthermore, the sensors’ readings may differ from the real ones concerning calibration or precision. There may be a construction feature that causes the car to behave differently in reality than in a simulation. Even the brightest data scientist is unable to predict all the possible scenarios. What would happen if a bird started to peck at the camera? Or if a car encountered a boy dressed as Superman pretending to fly? Or, more plausibly, after a collision, if there were an oil stain that looked exactly like a puddle, but would obviously have an entirely different effect on the tires’ grip of the road?
To address these challenges, data scientists randomize the data and the training environment to let the model gather more varied experiences. The model will learn how to control the car in changing weather and lighting conditions. By changing the camera and sensor settings, a neural network will gain enough experience to handle the differences or any changes that may occur when the model is being used.

Fighting the gap every lap

Using a simulated environment is one effective way of evaluating a model, but a pronounced reality gap still remains. To acquire better information (and also to have some fun on the job), data scientists evaluate their neural networks by launching them in small-scale models. There is currently an interesting Formula 1/10 autonomous car racing competition being held. Designing the software to control cars and compete against other teams is a challenging (yet fun) way to evaluate models. Small-scale models are tested on tracks with angles and long straights that are perfect for acceleration. Although the cars aren’t driven by humans, the team provides full-time technical assistance and puts the car back on track when it falters.
It’s also a great way to impress the jury in a similar manner to what Lawrence Sperry did more than a hundred years ago!
If this sounds interesting to you, deepsense.ai has recently launched a deep learning workshop where participants will learn to create models to control these small cars. The training will be conducted in cooperation with one of world champion Formula 1/10 racers.
The text was prepared in cooperation with Krzysztof Galias, deepsense.ai data scientist.

AI Monthly digest #1 – AI stock trading & Kaggle record

AI Monthly digest #1 – AI stock trading & Kaggle record

/in , , , /by and

AI-based stock trading, a record-breaking competition on Kaggle and more stories cherry-picked from all the interesting ML- and AI-related news from September. Right here in the AI Monthly Digest.

The Digest gathers machine learning and AI news to spot the most important and interesting events and developments of the past month. The five events below were curated from last month’s events and chosen by Arkadiusz Nowaczyński and Konrad Budek from deepsense.ai team.

Deep learning takes a deep dive into the stock market

Deep reinforcement learning can be applied as a complete AI solution for algorithmic trading.
The authors of “Deep Reinforcement Learning in Portfolio Management” set out to determine whether methods derived primarily for playing Atari games and continuous control would work on the stock market. The algorithm they used, called deep deterministic policy gradient (DDPG), returned promising results in an offline backtest.
The second paper, “Deep Reinforcement Learning in High Frequency Trading,” provides convincing arguments about why AI stock trading is suitable for trading in a timescale below 1 second (High Frequency Trading). The authors did a solid evaluation of their approach with a few noteworthy tips:

  1. Online learning at test time makes it possible to maintain high accuracy over time;
  2. A small neural network is enough for this problem, meaning AI for trading can be developed on laptops;
  3. Predicting the next 100 ticks from the last 500 ticks works best for them.

Progress remains to be made and questions to be answered. Does this algorithm work when deployed on the real market? How much money can you actually make with it? The lack of answers is certainly intriguing, as is the fact that algorithmic trading may soon be powered mostly by Deep RL, if it’s not already. We think that the potential financial reward will push people to develop further breakthroughs in AI. After all, setting high scores in Atari games isn’t as satisfying as having supersmart AI earning you gobs of money.
For those of you who are new to the topic we recommend a good nice-to-read introduction.

A record-breaking Kaggle competition

Over 8500 data scientists on no fewer than 7000 teams took part in the Kaggle Home Credit Default Risk evaluation record-breaking competition. The goal of the competition was to predict the risk of giving a loan to a particular customer. The teams were provided with rich datasets containing historical and transactional data on the customer’s behavior.

Related:  Keras vs. PyTorch: Alien vs. Predator recognition with transfer learning

Perfectly designed, the competition attracted teams from far and wide, mostly thanks to the outstanding dataset. It allowed the teams to harvest insights and play with data in often surprising ways. Looking to tune up their models and further polish their skills, participants engaged in discussions and peer-reviews long after the competition had ended.
deepsense.ai took part in the competition, with Paweł Godula leading a team that took 5th place overall and finished first on the public leaderboard.

Volvo trucks introduce Vera, the cabless truck

According to PwC data, by 2030 the transport sector will require 138 million fewer cars in Europe and the US, mostly thanks to the rise of autonomous vehicles and the development of new business models. What’s more, it is predicted that by 2030 autonomous vehicles will be driving 40% of all miles driven.
As a proof of concept, Volvo has brought out Vera, the cabless truck to be used in short-haul transportation at logistics centres or ports. With the fleet of vehicles able to communicate and be supervised by a cloud-based management system, the truck is an interesting glimpse of the driverless future.

DARPA announced $2 billion investment in AI

At it’s 60th anniversary conference, the DARPA (Defense Advanced Research Projects Agency) announced that it is going to invest $2 billion in artificial intelligence. The agency is known for developing cutting-edge technology, be it ARPANET, which later evolved into the Internet, or the Aspen Movie Map, which was among the predecessors of Google Street View.
According to John Everrett (via CNNMoney), the deputy director of DARPA’s Information Innovation Office, the agency’s investment is intended to accelerate the development of AI from 20 years down to five years.
DARPA’s investment is not the first a government has made in AI. The most notable example comes from the United Arab Emirates, which has appointed an AI minister.

NIPS conference sold out in less than 13 minutes

NIPS, hosted in Montreal, Canada, is currently the most important machine learning and AI research conference in the world. Initially held as an interdisciplinary meeting of experts interested in sharing their knowledge on neural networks, it has evolved into the machine learning meeting with thousands of papers sent for review. It is also a place to run competitions with the “Learning to run” in 2017 as an example.

Related:  Expert augmented reinforcement learning - agents of Montezuma’s Revenge

In 2017, the tickets sold out in two weeks, a relative eternity compared to the rock concert-like 12 minutes and 38 seconds they flew out in this year. Tickets for last year’s Comic-Con, one of the world’s most beloved pop culture events, sold out in a bit more than an hour.
So, when it comes to selling tickets, Marvel superheroes would appear to have nothing on machine learning. This year’s NIPS conference will feature Henryk Michalewski, visiting professor at Oxford University and a researcher at deepsense.ai, as a co-author of “Reinforcement Learning of Theorem Proving” paper.

Summary

September has clearly shown that AI is one of the most dominant trends in modern tech. Selling out venues faster than pop culture events goes a long way to proving that a scientific conference, or at least this one, can be as exciting as a concert or show – so long as it’s about Artificial Intelligence.

How machine learning improves business efficiency - five practical examples

How machine learning improves business efficiency – five practical examples

/in /by

Deloitte estimates that in 2021 enterprise spending on artificial intelligence and machine learning projects will reach 57 billion dollars, four times more than in 2017. These technologies are now in every day use, and not only among innovation leaders.

Thanks to digitalization of business processes, organizations command ever greater amounts of data, which, with the help of machine learning, can be used to automate work. At the same time, spending on the following five areas can be limited:

  1. maintenance, thanks to reduced energy consumption
  2. payroll costs, thanks to task automation
  3. raw material and quality assurance costs, thanks to the automation and tightening of quality control
  4. equipment and machinery costs, thanks to the automation of control systems for operations and maintenance
  5. operating costs including marketing and sales

More and more of the business community is catching on to the savings they can harness with artificial intelligence. The evidence for this is clear from the steps individual enterprises are taking, as well as the development of numerous machine learning business examples and the entire ecosystem of companies offering products based on these technologies and support in implementing them.

1. Reducing the costs associated with maintaining and using the energy

As a system’s complexity grows, so too does the challenge of supervising it. Click To Tweet
Consider, for example, the cooling of large server rooms. In terms of energy consumption and CO2 generation, the ICT sector (communication technologies, including telecommunications and IT services) produces two percent of global emissions, which is on a par with the airlines. To reduce its electricity expenses, Google decided to entrust energy management in one of its server rooms to AI, which “learned” the structure of the center and reduced cooling costs by 40 percent.
No new equipment was needed – it was enough to develop new software that leveraged AI. Ultimately, the system is going to be used in all Google server rooms. The British national energy supplier National Grid has also expressed interested in the solution.

Related:  Five trends for business to surf the big data wave

2. Reduction of human costs through automation

Machine learning enables the automation of repetitive, often time-consuming activities, freeing up the teams that had been doing them to take up more profitable tasks. We produced a program for the international research company Nielsen that could find, read and save in a database the composition of the a product’s contents using only a photo of its packaging. This shortened the working time from several minutes dedicated to manually rewriting the composition from the label, to the few seconds required to take a picture of the packaging.
If you need more convincing, consider these figures: If a company employs 46,000 people, helping even half of them save five minutes a day translates into 314 full-time positions each day.
Machine learning business examples

3. Predictive maintenance 4.0 – optimizing machine maintenance costs

Because any hardware failure involves both repair costs and production downtime, what company wouldn’t look for tools that can predict failures and prevent them? Another solution deepsense.ai has prepared, this time for a manufacturer, used data from sensors mounted on machines.
By reviewing and analyzing the signals, the solution can predict upcoming failures up to two weeks before they occur. Another example of predictive maintenance comes thanks to the OneWatt company, which tests the sounds issued by industrial machines. AI steps in where the human ear would be helpless: It detect changes in the sounds the machines produce to predict potential failures.

Related:  Three reasons why data analysts make the perfect data scientists

4. Quality control – fewer mistakes with machine learning

In many industries, quality control comes with huge costs. For semiconductor manufacturers, huge means up to 30 percent of costs. Automating quality control with image recognition tools increases the percentage of defects detected by up to 90 percent.
Unlike automated systems, machine learning-based vision systems can continuously evolve and adapt to new product specifications. Fujitsu implemented a system that both catches defective products and prepares each one for automated assembly at the next production stage. Applying machine learning, the system not only automatically recognizes the parts of the machine but also assesses their compliance with standards in more than 97 percent of cases.
Machine learning business examples

5. The power of data in sales, marketing and customer service

Machine learning is able to process data sets faster and more efficiently than even the most expert analysts. This makes it possible to constantly analyze what is happening, for example, in the company’s sales or transaction system, and also to regularly monitor customer activity. To understand just how beneficial machine learning can be here, consider the customer loyalty survey.
Only one in every 26 clients expresses their dissatisfaction before looking to the competition for what they need. Data science can help you capture the behavior patterns of a dissatisfied customer and react in advance.
Machine learning business examples
Using data more effectively benefits not only business, but the whole of society. A solution developed by deepsense.ai for the city of Portland, Oregon enabled the police to predict in which parts of the city crime would take place.

Machine learning in practice

Machine learning is giving enterprise more opportunities to look for savings and generate additional revenue. AI helps people accomplish complex tasks that under normal conditions would overwhelm them, so great is their complexity. Machine learning also makes it possible to automate activities that, though repetitive and schematic, require maximum focus, so employee productivity tends to fall quickly. AI helps people do their work more effectively and devote more energy to those activities that bring the most value.

When predictive analytics in football fall short (an example)

When predictive analytics in football fall short (an example)

/in /by

According to predictions done by Soccerbot 3000, the AI-powered prediction machine, Germany should face Brazil in the finals of the World Cup in Russia – or should have, that is. Then the unthinkable happened.

The short explanation for those not interested in the football matches being played in Russia: the German team – the same one Goldman Sachs picked as the probable world champ – failed to get out of its group for the first time in 80 years. And when the German team was vanquished by South Korea and its brilliant Son Heung-Min, predictions were proved wrong. Not much later, the always ballyhooed Brazilian team was knocked just as far out of the tournament, which is to say, all the way. Indeed, Mr. Neymar and peers, following legends Cristiano Ronaldo and Leo Messi, were sent packing before the semi-finals.
The Financial Times pointed out that Soccerbot 3000 used “200,000 models” that generated “1,000,000 possible evolutions of the tournament”. According to its prediction, this year’s Cup should have gone to Brazil by a nose, or a toe as it were. The conclusion that Machine Learning is still ineffective in predictions seems obvious, but it would be severely biased.

It wasn’t only the model that was surprised

In the detailed report on the current World Cup and among generated vast majority have shown Brazil, France, and Germany were forecast to lead, with 18.5%, 11.3%, and 10.7% chances of bringing home the cup, respectively. None of the other teams garnered more than 10%.
The models used historical data about team characteristics, individual players and recent team performance. The model later learned the correlation between these metrics and the teams’ performance based on World Cup data since 2005. Is that a massive amount of data? Indeed it is. But that’s hardly the entire issue.
The model was unable to predict the weather, player health or the atmosphere prevailing in each team. Football, like any other game, consists of many more variables than researchers are able to predict and insert into a model. Of the one million scenarios it produced, the model predicted almost 200,000 scenarios when Germany didn’t reach the round of 16.
That it didn’t happen shocked the world, not only people who trusted the AI to predict the outcome. Even the famous Gary Lineker, who said that “Football is a simple game – twenty-two men chase a ball for 90 minutes and at the end, the Germans always win.” after Germany have beaten England in Italy in 1990 updated his famous quote.

Racing with probability

The key role machine learning models play is in reducing the randomness of choices based on data processing. The level of accuracy applied to be used in production is highly dependant on the purpose it was designed for.
A fraud detection model that was 80% accurate would never be used in a bank or any other institution. The 20% of the fraud it didn’t catch would be nothing short of a disaster for such an institution.
On the other hand, a model that could return that same 80% processing investment opportunities would earn millions of dollars. Warren Buffet may have missed the investment opportunity in Google and Amazon, but that doesn’t make him an unreliable investor.
When predictive analytics in football fall short (an example)
Considering its 81,25% accuracy in the group phase, the model would be quite reliable as an advisor, even if it was unable to read opinions, use social media, leaked information or just read the news just before each match to make corrections.
When a company has access to more reliable data or even provides all the data possible, the accuracy rises. This can be seen in visual quality control or recognizing diabetic retinopathy from photos. Predicting the outcome of sporting events is a much different business.
Even the Goldman Sachs analysts behind the model cautioned against seeing it as an oracle. In any case, however many analyses or however much data science gets done, the World Cup will be exciting to watch.