2020 - the year in review

2020: the year in review

/in /by

Tomasz Kułakowski, CEO, talks about how deepsense.ai entered the new decade and dealt with the new business reality.

How do you rate the past year?

I won’t be original if I say that the past year has been full of challenges. From a global perspective, we can talk about the crisis, the interruption of supply chains and paralysis in many segments of the economy. However, the market uncertainty affected the IT industry to a lesser extent. Analysts predict a continuous increase in the value of services related to cutting-edge AI-focused technologies.

2020 has definitely taught us humility and perseverance. At the same time, we made sure that, as a company, we are moving in the right direction. Our business model, persistence in achieving goals and world-class expertise have together enabled us to proactively react to enormous market changes and flexibly approach the new business reality taking shape.

What are you most proud from the past year?

Unquestionably, I can say I’m most proud of the great responsibility and commitment all of deepsense.ai’s team members have shown. I have never doubted our team’s spirit, knowledge and competence. However, only crisis situations such as the Covid-19 pandemic are the real test. Today, I’m totally convinced that nothing is impossible for us.

My optimism is confirmed by the rankings. Last year, Robert Bogucki, CTO of deepsense.ai, was named to the prestigious Forbes 30 Under 30 list of young visionaries. Moreover, our company was awarded an honorable mention in the big data ecosystem of the most impactful companies on the insideBIGDATA IMPACT 50 List for Q4 2020.

It is said that if a company survives in the market for 5 years, it will survive anything. We faced Covid-19 in our 6th year of operation. We not only survived, but as an organization also grew up at a much faster pace.

What were the biggest challenges for deepsense.ai in 2020?

The biggest challenge for us was convincing our clients that despite the uncertain situation on the market, it is worth continuing work on AI projects, as they will provide real business value.

It was also crucial for us to maintain team spirit and effectively work almost 100% remotely, without disrupting the company’s high level of customer service.

Continuing to invest in projects developed within the R&D Hub was also of vital importance. Many of our clients appreciated the fact that we continued our research and drew inspiration from a changing world.

What are deepsense.ai’s plans for 2021?

The year 2020 has confirmed that AI-focused solutions are the natural next stage in the development of the modern world. Improvements based on computer vision, natural language processing and predictive analytics enable the automation and optimization of various business processes. These will accelerate the pace of recovery from the crisis. Many of our clients have told me that if they had implemented AI solutions before the pandemic hit, the crisis would have impacted their business much less severely. Such insights allow us to set ambitious goals for 2021.

Approximately 85% of our revenue currently comes from the US and Western Europe, and we plan to expand our presence on these markets. We will also continuously invest in research on the development and commercialization of new technologies within the R&D Hub. Last but not least, we want to maintain our leading position in reinforcement learning research.

I hope the new year will bring deepsense.ai the opportunity to unleash the energy we’ve accumulated in 2020. That will mean finding more outlets for our creative and analytical powers in industry and research. It will also mean redoubling our energy and efforts to take on new opportunities the new year brings.

deepsense.ai paper to be featured at ICLR spotlight

deepsense.ai paper to be featured at ICLR spotlight

/in /by

“Model-Based Reinforcement Learning for Atari,” a research paper written by a consortium of authors, has been chosen for a spotlight presentation at the International Conference on Learning Representation. Only 5% of the papers submitted receive this distinction. Two of the leading authors, Piotr Miłoś and Błażej Osiński, are both researchers and senior data scientists at deepsense.ai.

The paper describes research conducted in cooperation with Google Brain, the University of Warsaw and the University of Illinois at Urbana-Champaign.

The high-level goal described in the research is how to enable an AI model to learn useful behaviors by interacting with its environment. In this case, the environment was Atari games, which offer tasks that are simple yet challenging enough to test the research concept. The results of the experiment can be extrapolated to more challenging tasks in increasingly complex environments. The model the deepsense.ai researchers designed learned how to play the games thanks to an extensive chain of trial and error.

An extension of the research might be a robot acquiring skills while moving objects, such as putting dishes in a dishwasher or refrigerator. Such interactions can be costly – the robot will drop any number of dishes along the way, so it takes time and resources for it to train itself thoroughly enough to learn to carry out its tasks efficiently.

To tackle the challenge of expensive interactions, we increase the system’s learning efficiency by endowing it with a predictive module that can imagine the consequences of actions without actually executing them. The module is trained along with this skill and it is used to perform short inner simulations. These simulations substitute most of the interactions normally required by reinforcement learning, resulting in better sample efficiency.

Details about the research are available in the full arXiv preprint, while the press release and an article about artificial imagination can be found on deepsense.ai’s blog.

“We are proud and excited to see our work recognized at one of the most renowned scientific conferences in the world,”
– says Piotr Miłoś, senior data scientist at deepsense.ai and a professor at the Polish Academy of Science.

Reinforcement learning in spotlight

“Spotlight” is a short oral presentation reserved for papers of particular distinction. With only 5% of submitted papers honored this way, the privilege comes amidst our work gaining critical attention in the AI research community. The arXiv paper has been cited more than 50 times in just 10 months, a testament to its significance for other researchers in the field.

This year’s conference, the eighth, will be held in Addis Ababa, Ethiopia, and is hoped to encourage researchers from Africa and the Middle East to take part. Despite its short history, the conference is currently among the most important deep learning-related scientific events  in the world. The conference publication is recognized on Google scholar as the 42nd most important scientific publication globally.

The conference provides an interesting insight into the development of Artificial Intelligence and is a great way to keep abreast of the latest trends. deepsense.ai has recently used machine learning tools to review the papers submitted to the conference. The analysis showed that reinforcement learning is among the most significant and recurrent trends in machine learning today. Further information on trending topics on the ICLR conferences can be found in deepsense.ai’s analysis.

“Presenting our research on spotlight is a great honor and proof that our contribution to the machine learning community and development is recognized,”
– concludes Błażej Osiński, a senior data scientist and deepsense.ai researcher involved in the project.

This success is the result of deepsense.ai’s strong commitment to R&D. Just a month ago two other papers were presented at NeurIPS workshops, another top-tier AI conference. These included “Developmentally motivated emergence of compositional communication via template transfer” and “Simulation-based reinforcement learning for real-world autonomous driving”.

The full list of paper authors: Lukasz Kaiser, Mohammad Babaeizadeh, Piotr Milos, Blazej Osinski, Roy H Campbell, Konrad Czechowski, Dumitru Erhan, Chelsea Finn, Piotr Kozakowski, Sergey Levine, Afroz Mohiuddin, Ryan Sepassi, George Tucker, Henryk Michalewski

###

About deepsense.ai:

brand-manager-at-deepsense

deepsense.ai research presented at NeurIPS 2020

deepsense.ai research presented at NeurIPS 2020

/in /by

It is the second time deepsense.ai researchers have presented at the NeurIPS conference. In the most recent installment of the world’s most closely watched AI conference, Piotr Miłoś described CRTS research during the Machine Learning for Autonomous Driving workshop. The presentation continued deepsense.ai’s fruitful cooperation with Volkswagen on training autonomous vehicles in a simulated environment.

The R&D project was conducted by a team of researchers including deepsense.ai’s Miłoś and Michał Martyniak, as well as Błażej Osiński, Adam Jakubowski, Paweł Zięcina, Christopher Galias, Henryk Michalewski together with Volkswagen researchers Silviu Homoceanu and Antonina Breuer.

“During this stage of the project we developed a port of real-life traffic footage into the CARL simulator, which we called CARLA Realistic Traffic Scenarios (CRTS). The aim was to provide a training and evaluation ground environment for obtaining better driving policies. We believed that an agent developed within the CRTS scenarios would behave more naturally when deployed, due to the simple fact that it has had a chance to encounter more realistic situations while training,”
explained Miłoś, who in addition to working as a researcher at deepsense.ai is a professor at the Polish Academy of Sciences.

The collaborative research has led to a number of noteworthy breakthroughs. The artificial agent obtained with reinforcement learning has become “more dynamic” than the reference human driver. That it drives faster is one example of how. Being able to use simulation for such experiments offers crucial advantages, including being less expensive and less time-consuming. Over the combined experiments, the team generated over 100 years of simulated driving experience.

The observations that resulted from the experiment bring valuable benefits both for scientific and commercial projects focused on real-life developments of autonomous cars. Details of the research are described in the team’s paper as well as on the project’s website.

deepsense.ai and Volkswagen deliver the breakthrough in autonomous car research

deepsense.ai and Volkswagen deliver breakthrough in autonomous car research

/in /by

deepsense.ai and Volkswagen show that an autonomous vehicle trained entirely in simulation can drive in the real world. The team transferred its neural network into a real car, providing it with over 100 years’ driving experience before the real-world engine had ever started.

Researchers from deepsense.ai recently presented their work on training an autonomous vehicle in a simulated environment and then transferring the model into a real-world vehicle. The team presented their paper during a workshop at the 2019 Neural Information Processing System conference in Vancouver, BC.

“Moving the neural network policy from the simulator to reality was a breakthrough and the heart of the experiment. We have run multiple training sessions in the simulated environment, but even the most sophisticated simulator delivers different experiences than what a car encounters in the real world. This is known as the sim-to-real gap. Our algorithm has learned to bridge this gap,” explains Piotr Miłoś, deepsense.ai researcher and a professor at the Polish Academy of Sciences. „That we actually rode in a car controlled by a neural network proves that reinforcement learning-powered training is a promising direction for autonomous vehicle research overall.”

The experiment was conducted by a team of researchers including deepsense.ai’s Błażej Osiński, Adam Jakubowski, Piotr Miłoś, Paweł Zięcina and Christopher Galias, Volkswagen researcher Silviu Homoceanu and University of Warsaw professor Henryk Michalewski. The team not only trained a model that controls a car in a simulated environment but also executed test drives in a real car at Volkswagen’s testing facility. The car navigated through real streets and crossroads, performing all sorts of real-world driving maneuvers.

Over 100 years of driving experience

The technique the deepsense.ai-Volkswagen team used was innovative for two reasons. First, the neural network was trained in a simulated environment and later transferred to a real car. Second, the neural network was trained using a reinforcement learning paradigm.

Being able to use simulation offers crucial advantages. First, it is cheaper. In their combined experiments, the team generated some 100 years of simulated driving experience. Racking up so much experience in a real car isn’t feasible. Training in a simulator can also be done much faster. The agent can experience all manner of danger, from simple rain to deadly extreme weather, accidents and full-blown crashes, and learn to navigate or avoid them. Subjecting an actual driver to such hazards would be prohibitively complicated, time-consuming and ethically unacceptable. What’s more, extreme scenarios in the real world are relatively uncommon but can be brought on in a simulator at the snap of the fingers.

“It was exciting to see that our novel approach worked so well. By further exploring this path we can deliver more reliable and flexible models to control autonomous vehicles,” explains Błażej Osiński. “We tested our technique on cars, but it can be further explored for other applications.”

Shaping the behavior

Reinforcement learning allows a model to shape its own behavior by interacting with the environment and receiving rewards and penalties as it goes. The model’s goal is, ultimately, to maximize the rewards it receives while avoiding penalties. An autonomous car receives points for driving safely and complying with the traffic laws.

Reinforcement learning delivers a more flexible model that can adjust itself to a changing environment and react accordingly in new situations. Simulating every road condition that can occur is impossible, but shaping a set of guidelines like “avoid hitting objects” or “protect passenger from any and all harm” is not. Ensuring the model sticks to the guidelines makes it more reliable, including in less common situations it will eventually face on the road.

Technical details of the research are described in the team’s white paper, available on arxiv. At the 2019 Neural Information Processing Systems conference in Vancouver, BC,  over ten thousand participants took part in sessions and workshops on artificial intelligence. The team presented its paper during the workshop Machine Learning for Autonomous Driving (https://ml4ad.github.io/) organized by researchers from UC Berkeley, Carnegie Mellon University, Columbia University, and Waymo.

###

About deepsense.ai:

brand-manager-at-deepsense

deepsense.ai names Paweł Osterreicher and Bartek Klimczak to Management Board and expands its executive team in the United States

deepsense.ai names Paweł Osterreicher and Bartek Klimczak to Management Board and expands its executive team in the United States

/in /by

AI business application leader deepsense.ai has appointed two new members to its Management Board. From November, Pawel Osterreicher will assume the role of Chief Revenue Officer (CRO) while Bartek Klimczak will become Chief Operating Officer (COO). Rounding out the executive moves, Andy Thurai joined the company as Head of US Operations.

“I am proud and excited to welcome such strong players as Paweł and Bartek to the management board,” says deepsense.ai CEO Tomasz Kułakowski. “The company is currently witnessing rapid growth and immense geographical expansion. Our key challenge is to reach the ambitious goal of securing growth. Seasoned experts like Bartek and Paweł will support deepsense.ai’s position as a global AI leader.”

Osterreicher has been deepsense.ai’s Director of Sales and Strategy since April 2018. He came over from Boston Consulting Group (BCG), where he headed up projects in the financial and consumer/retail sectors, focusing his efforts on strategy, transformation, the building out of digitalization and implementation of advanced analytics. Prior to that, Paweł worked with the A.T. Kearney and Santos company (Australia).

“I am very optimistic about the future growth of the company and I am very much looking forward to coordinating deepsense.ai’s sales efforts,” says Osterreicher. “Some of the recent successes deepsense.ai has had, and which I contributed to, involved acquiring multiple key business customers and partners. We have also successfully grown our competencies thanks to projects that have pushed the boundaries of what is possible in AI. I am convinced that deepsense.ai’s potential, comprising both its employees competencies’ and its acquired business know-how, is unique. Further scale-up of the company’s activities on a global level will be a challenge – a pleasant and very interesting one.”

Bartek Klimczak joined deepsense.ai in mid-2019 as a Managing Director. Prior to that, he spent seven years at The Boston Consulting Group (most recently as Principal), focusing on a broad spectrum of strategic and operational projects for leading global players in industrial goods and processing sectors as well as energy and the global oil & gas sector. Prior to BCG, Bartek spent another 7 years in investment banking in the CEE (the CAG & KBC Group), where he primarily brokered M&A deals.

“I see massive growth potential and unique opportunity for deepsense.ai, especially in industries driven by Industry 4.0 adoption of AI/ML across the entire value chain, from supply chain optimization, through machine vision-based automatic quality control to predictive maintenance, machine auto-diagnostics and work safety smart intelligence. All of this will bring our clients’ businesses in Poland and globally to entirely new levels in terms of operating efficiency and profitability. By leveraging our unique capabilities and proprietary AI/ML technologies, we are able to effectively target even the most challenging and custom problems our clients may face throughout their production processes. We want our focus aimed primarily at areas where properly designed AI can bring significant results. That will be our wheelhouse moving forward,”
– explains Klimczak.

The promotions are the latest in a string of moves deepsense.ai has made to expand its executive team. The company has also recently tapped Andy Thurai to head up its US Operations and help the company expand its footprint in the United States, the most promising market for its cutting edge technologies.
Mr. Thurai worked as a chief strategist for IBM and an Emerging Technology Strategist for Oracle Cloud Infrastructure. He has also been a field CTO with Intel and BMC, and has extensive experience as a consultant, technology advisor and mentor.

“deepsense.ai combines technical proficiency with a deep understanding of business use cases, being one of the very few companies ready to solve problems once thought were unsolvable. I am proud to support the company with my experience,”
– says Mr Thurai.

The company also welcomed Nina Simosko to its Advisory Board. A former CEO of the NTT Innovation Institute, the prestigious Silicon Valley-based innovation center for NTT Group, Ms. Simosko is a well-known Silicon Valley technology executive with senior management experience in both global Fortune 500 companies and innovative startups.

###

About deepsense.ai:

brand-manager-at-deepsense

deepsense.ai names Andy Thurai head of US operations

deepsense.ai names Andy Thurai Head of US Operations

/in /by

Leading machine learning and artificial intelligence-based solutions provider deepsense.ai has named  Andy Thurai as head of its US Operations. Mr. Thurai gained broad experience at tech companies including IBM, Oracle, BMC and Intel, where he was responsible for product strategy, thought leadership and business development.

“I am happy and proud to invite Andy on board,” said Tomasz Kułakowski, deepsense.ai’s Chief Executive Officer. “Andy is a seasoned strategist with a strong tech background and thorough understanding of the cutting-edge technology we are dealing with at deepsense.ai. His involvement will lift deepsense.ai into a new era and further support US operations.”

Mr. Thurai worked as a chief strategist for IBM and an Emerging Technology Strategist for Oracle Cloud Infrastructure. He was a field CTO with Intel and BMC. He also has extensive experience as a consultant, technology advisor and mentor.

“deepsense.ai is a technological trail-blazer and one of the most innovative companies in the world, delivering outstanding results and operating on the cutting-edge tech,” Mr Thurai said. “I am truly excited to join the company and help change the world for the better with my experience. It is going to be a great adventure!”

###

About deepsense.ai:

brand-manager-at-deepsense

deepsense.ai names Nina Simosko to advisory board

deepsense.ai names Nina Simosko to advisory board

/in /by

deepsense.ai, a leading AI-based end-to-end solutions provider with the focus on computer vision, predictive analytics and natural language processing, today announced that Nina Simosko has been appointed to the company’s advisory board. Ms. Simosko, formerly the CEO of NTT Innovation Institute, the prestigious Silicon Valley-based innovation center for NTT Group, one of the world’s largest ICT companies, is a well-known Silicon Valley technology executive with senior management experience in both global Fortune 500 companies and innovative startups.

“I am very pleased that Nina is joining our advisory board,” stated Tomasz Kulakowski, deepsense.ai’s chief executive officer. “We look forward to benefiting from her strong background in cutting-edge technologies, supporting our mission to improve people’s lives with machine learning-powered solutions. We welcome her strategic input and the combination of leadership, experience and industry knowledge she brings to deepsense.ai as the company transitions to the next phase of growth.”

Previous to NTT i3, Ms. Simosko was responsible for leading the creation and execution of Nike Technology strategy, planning and operations world-wide. Prior to that, she was Senior Vice President of SAP’s Global Premier Customer Network (PCN). At SAP, she led both the PCN Center of Excellence and SAP’s Global Executive Advisory Board. During her eight-year tenure, she was a part of SAP’s Global Ecosystem & Partner Group which was charged with continuing to build and enable an open ecosystem of software, service and technology partners together with SAP’s communities of innovation. Ms. Simosko currently sits on the Advisory board at Santa.io, AppOrchid and Reflektion and she has also been a member of the advisory boards at Appcelerator and Taulia.

“I am excited to join the company’s advisory board”, stated Ms. Simosko. “deepsense.ai is potentially solving one of the biggest challenges companies currently tackle in identifying, analyzing and solving problems with AI-powered solutions across multiple industries. I highly appreciate the company’s focus on research and their ability to deliver meaningful business results to their customers.”
– Nina Simosko

###

About deepsense.ai:

brand-manager-at-deepsense

deepsense.ai launches Research and Development Hub

deepsense.ai launches Research and Development Hub

/in /by

“There is no development without research. To make our efforts in exploring new lands in machine learning more structured, deepsense.ai has launched its Research and Development Hub,” explains Tomasz Kułakowski, CEO of deepsense.ai.

deepsense.ai’s Research and Development Hub consists of four full-time researchers supported by a technical team. Supporting deepsense.ai’s business goals, the Hub has a fixed research budget funding projects that bridge the gap between enterprise and academia.

“Sometimes AI research projects seem distant from our everyday experience. But it’s actually just the opposite. The research we are doing today will power the autonomous cars and sophisticated prosthetic limbs of the future. Projects based on Atari games give us benchmarks on applying new techniques in computational-heavy projects done in more heavyweight simulators.”
– Tomasz Kułakowski, CEO of deepsense.ai.

The team has conducted numerous successful research projects in cooperation with global leaders including Google Brain, Intel, a leading car manufacturer as well as top universities and scientific institutions around the world.

Key projects

deepsense.ai is currently focused on model-based reinforcement learning and transferring models from a simulated environment to the real world (sim2real). Reinforcement learning is a driving technology behind the large strides that have been made in AI, including recent success in Go, Chess, Dota 2 and StarCraft II.

Model-based reinforcement learning and learning in simulation address the challenge of collecting voluminous real-world data. That challenge does not exist in board games such as Go, Chess and e-sports, where one can collect an equivalent of millions of years of human gameplay and learn from such extensive datasets. It is, however, a major obstacle for real-world applications of reinforcement learning such as robotics.

In a recent project – Model-Based Reinforcement Learning for Atari – deepsense.ai set out to improve the quality of video models used in model-based reinforcement learning. Neural networks developed in this project are the state of the art in model-based reinforcement learning. The researchers were invited to present their work at Oxford University, Google Brain and the research company DeepMind.

Another project consists of training in Unreal Engine 4 and deployment on a real car. The objective of this research is to assess the feasibility of training a fully functional autonomous driving agent in simulation with only a minimal amount of real-world data. The project is being conducted in cooperation with a leading global car manufacturer.

“Long-term investment in research work is a distinguishing feature of deepsense.ai. Over the past three years, we have written a number of papers and presented at important ML venues including NeurIPS. All of this is done with a firm conviction in mind: machine learning is changing and developing so fast that today’s research can be tomorrow’s gold standard,”
– Henryk Michalewski, head of deepsense.ai’s R&D team.

More of the research and development projects deepsense.ai has undertaken can be found on its recently published R&D Hub subsite.

Delivering the future

Apart from scientific work, the R&D Hub will make AI research more accessible and comprehensible for non-engineers and people who are unfamiliar with data science.
“Our aim is to provide readers with as much information on our work as possible. Research should not be a hermetic game. It is about building a beneficial future,” comments Kułakowski. “I am proud that deepsense.ai is contributing to the development of this fascinating discipline.”

###

About deepsense.ai:

brand-manager-at-deepsense

Shadows of customers on the wall – key takeaways from the “AI in e-commerce” business breakfast

/in /by

Global e-commerce is among the fastest growing industries globally, experiencing 18% growth in 2018. Worldwide, consumers purchased $2.86 trillion worth of e-goods in 2018, compared to $2.43 trillion in 2017.

Because digital commerce is data-driven, the industry is ripe territory for AI. However, lack of knowledge and uncertainty remain the most prominent obstacles to this technology gaining a stronger foothold. To address these obstacles, deepsense.ai and Google Cloud co-organized a business breakfast to discuss the challenges and opportunities and share their remarks on artificial intelligence (AI) in e-commerce. Joining Google and deepsense.ai were experts from BeeCommerce.pl, Sotrender, and iProspect, all companies deliver sophisticated tools for digital business.

Plato’s data cave

“When it comes to building AI applications, it’s all about the data,” said Paweł Osterreicher, Director of Strategy & Business Development at deepsense.ai, during his presentation. He pointed out that the simplest analytics in smaller businesses can be done within an Excel spreadsheet or pen and paper. Preparing a simple segmentation within a client group or spotting best-performing products don’t pose a huge challenge. But those are only the tip of the iceberg. “The more sophisticated insights we gain, the more complicated the task becomes. And that’s where specialized software comes in,” he said.

“The greatest challenge is a lack of flexibility. There is no jack-of-all-trades among the popular tools, and each has its limitations. The problem is when a tool doesn’t fit a company’s needs. And, to be honest, that’s a common situation,” Osterreicher continued. Companies thus often need to tweak the tools at their disposal to make them fit or get used to missing insights from their data.

“Most companies process only a fraction of their data and operate with only half the picture. They are like the prisoners in Plato’s cave, watching only the shadows customers cast on the wall, with no access to or true grasp of their real form.”

The only way to analyze data in a convenient and cost-effective way is to leverage machine learning models. Machines are able to effectively spot patterns even in seemingly insignificant details.

“Sometimes information about how long customers hover over a button or how they go about filling in an online form is a first step to obtaining meaningful information. The model is only as good as the data it was built on,” concludes Osterreicher.

Retail reinvented

In another presentation, Jakub Skuratowicz focused on the technical aspects of how companies use AI. There are numerous ways for companies to benefit from AI, be it building engagement, personalizing the user experience or detecting fraud.

Google’s expert showed a new application of image search for omnichannel commerce. First applied by the Nordstrom clothing company, the app-enabled users to take a photograph of an item and then search for it in the shop’s database. Thus, the customer could quickly buy the product online or check its availability.

“By using Google Cloud Platform-delivered machine learning tools, the company reached 95% accuracy in recognizing an item shown in a photograph”

AI also thrives in recommendation engines. “It was common to recommend the user another version of the product – a different size of a dress, for example. That’s pointless. Why would one need another of the same dress, only slightly bigger?” Skuratowicz asked. Instead, the AI-powered model recommended products that complemented the one that had been searched for, like sunglasses or a scarf to go with the dress.

Skuratowicz also showed how AI spots fraudulent transactions in international e-commerce. “Manual or semi-automatic checking can be effective, but machine learning makes it more scalable,”  he said. By applying AI-based solutions, the international logistics provider Pitney Bowes boosted the accuracy of its fraudulent transaction detection by 80% while reducing false-positives by 50%.

The mind barrier

The presentations were followed by a panel discussion on machine learning in e-commerce. As the panelists remarked, the AI-powered future of e-commerce is a challenge that not all companies are ready for.

In response to a question about the state of data-proficiency in e-commerce companies, Arkadiusz Wiśniewski, Director of Data and Technology at iProspect, had this to say:

“some data are easy to collect, while others provide a challenge, so we have an incomplete view. The legal situation in Europe poses an additional challenge, so it is better to focus on the data owned and make the best use of it.”

“Data-readiness depends to a great extent on company size. But most businesses lack the skills and data to effectively apply machine learning techniques,” agreed Jarosław Trybuchowicz, owner of beecommerce.pl.

The panelists agreed that the situation is hard even though data is becoming a commodity. “Sometimes the problem is the opposite. Despite having huge amounts of data, companies don’t get insights from it. They simply don’t know what questions to ask and what insights to look for,” added deepsense.ai’s Borys Sobiegraj.

The panelists likewise agreed that the key to success for enterprises employing machine learning is to know and properly organize their own data. Getting the data is the first challenge; “deciding what to do with it is a different story altogether,” said Jakub Nowacki, Lead Machine Learning Engineer at Sotrender. “Another challenge is extracting value that often lies in matching the data from different sources. If a company is unable to determine the impact of a sales campaign, then what is the purpose of analytics?” he added.

A question-answer session and networking time followed the discussion panel. The next business breakfast is planned for Q3.

###

About deepsense.ai:

brand-manager-at-deepsense

deepsense.ai-and-Google-Brain-design-artificial-imagination-for-reinforcement-learning

deepsense.ai and Google Brain design artificial imagination for reinforcement learning

/in /by

deepsense.ai, Google Brain, the University of Warsaw and the University of Illinois at Urbana-Champaign have concluded a collaborative research project, building neural networks that mimic a simulated environment and effectively enabling artificial intelligence to perform a simulation.

Designed models can be applied in text or video prediction and to solve the problems where mathematical description may be overly complicated.

Reinforcement learning (RL) enables a neural network to develop and hone its skill by interacting with its environment, which is usually a simulated one. The technique is predicted to shine in robotics and the building of autonomous vehicles.

The research team found a way to build a neural network that can mimic the signals that the RL agent usually gets from interacting with the environment. The neural network produces signals that usually would be taken from sensors, like images. “This is one of the important ideas of reinforcement learning. A recent survey done by DeepMind’s J.B. Hammrick provides a thorough account of analogies between model-based reinforcement learning and mental simulation as considered by cognitive science”, says Henryk Michalewski, the R&D Coordinator at deepsense.ai and a visiting professor in the Computer Science Department of the University of Oxford.

The challenges of education

In the learning process, the RL agent is rewarded for performing a task correctly and punished for making mistakes. Autonomous cars provide perhaps the best example–the agent is rewarded for safe driving and punished for collisions and speeding. The model seeks to maximize the rewards and minimize the punishments. Initially, all actions are random and the neural networks have yet to explore the possible ways to perform the tasks. It is thus possible that more than the first dozen rides will end with a collision with the nearest wall as the model incurs punishment and learns how to brake or avoid it.

This leads to the idea of conducting the trial and error process inside a simulator.

However, building a realistic simulator is a tedious problem for humans– says Błażej Osiński, a Senior Data Scientist at deepsense.ai. For example, Unreal Engine 4 has several million lines of code. The idea of our project is to let the neural network learn how to simulate the environment. A similar approach was suggested in a recent work from Yann LeCun’s lab, where neural networks were employed to simulate dense traffic.

The research team has built a neural network that emulates the Atari gaming environment, a popular training ground for reinforcement learning models. The network was able to create a version of the games Pong, Freeway and others that were nearly indistinguishable from Atari’s.

Ongoing research

The project was run by data scientists and researchers from deepsense.ai including Henryk Michalewski, Piotr Miłoś, Błażej Osiński and fellow researchers from Google Brain, the University of Warsaw and the University of Illinois at Urbana-Champaign. More detailed information about the research, outcomes and possible uses can be found in the Arxiv paper and a detailed blogpost about artificial imagination on deepsense.ai’s blog.

The research opens up interesting ways to apply neural networks in business settings. The first is to build an RL agent that can explore a highly complex environment and then emulate it for the needs of models that are to perform tasks within it. The best example may be teaching the model to emulate the real world in all of its complexity and unpredictability.

Another business application would be a video prediction tool that could deliver videos based on only a few frames, effectively reducing the effort animators and designers put into producing full-length videos.

We consider our research work an essential part of deepsense.ai and a key aspect of the company’s development. This time our researchers have really pushed the boundaries of knowledge with no marketing overstatements. The project brings fresh, new idea into AI research, a thing that is worth contributing into.– concludes Tomasz Kułakowski, CEO at deepsense.ai.

###

About deepsense.ai:

brand-manager-at-deepsense