deepsense.ai’s paper has been recognized by the General Chairs of the Conference on Robot Learning (CoRL) as one of 11 in the reinforcement learning and robotics category. It will be presented in November at Google’s headquarters in Mountain View.

In a recent research project, members of deepsense.ai’s machine learning team, Maciej Klimek, Henryk Michalewski and Piotr Miłoś, trained a robotic arm to grip a can of coke using reinforcement learning – in other words, through trial and error. The particular method the team developed was selected for presentation at the prestigious new robotics conference, CoRL 2017. The conference focuses on the intersection of robotics and machine learning and is being organized with the help of the International Foundation of Robotics Research (IFRR). deepsense.ai’s ‘Hierarchical Reinforcement Learning with Parameters’ will be featured along with 10 other standout archival papers on reinforcement learning and robotics selected for presentation at the conference and publication after the event.

The novelty of the approach presented in the paper comes from a clever division between simpler micro‑actions and general goals. The methods developed by deepsense.ai’s team are not limited to robotic arms, but could be used, for example, to train humanoid robots to combine single steps into a walk or a run.

According to Henryk Michalewski, Senior Data Scientist at deepsense.ai and an assistant professor at the University of Warsaw, “In recent years reinforcement learning has brought a number of striking new ideas to practical computer science. First, the DeepMind team managed to create bots which achieved superhuman performance in arcade games. Then, using a more sophisticated training methodology, they created a bot which beat the top human player at Go. A few weeks ago Open AI showed a bot that defeated the best human players at DOTA 2. At deepsense.ai, we also research reinforcement learning used in robotics, and we’re happy to have our work recognized on the international AI scene”.

Watch short videos with the robotic arm actions simulated during research:

###

About deepsense.ai:

Media contact:

deepsense.ai trademarks at boilerplate

The partnership carries over from last year’s collaboration, and will see deepsense.ai support one of this year’s main themes at AI World.

The AI World Conference and Expo will be held this year from December 11-13 in Boston, MA. deepsense.ai will again be a Strategic Deep Learning Partner, and will run a one-half day seminar Machine Learning and Deep Learning in the Enterprise on December 11. AI World will also host a webinar with deepsense.ai entitled “Investing Intelligently in AI Solutions“ on August 30, 2017 at 1:00pm EDT.

deepsense.ai builds its strategy around delivering Artificial Intelligence solutions for business and research organizations, with deep learning at the center. “For the last few years we’ve been building our machine learning and deep learning competencies constantly as we want to be the ones who help shape the world’s future”, says Tomasz Kulakowski, deepsense.ai CEO and co-founder. “We all already use AI in our daily lives and deepsense.ai is likewise participating – and hoping to push the technology forward. For example, the results of our work will help humans to more effectively diagnose diseases, including diabetic retinopathy and cervical cancer. In the coming few years, areas like fraud detection and failure prediction in manufacturing will become revolutionized by more and more common applications of deep neural networks being used to analyze various kinds of data. This will greatly increase the quality of services and products we then buy as consumers.”

In the December 11 deep learning seminar, deepsense.ai’s data scientists will explain what deep neural networks are, why they have become so popular in data science and how they can be used in business. Robert Bogucki, deepsense.ai Chief Science Officer, will also moderate a panel featuring heads of data science and innovation departments discussing how deep learning is disrupting their industries and critical lessons they have learned in deploying enterprise class machine learning applications.

“We are extremely pleased to be working again with deepsense.ai. Last year, deepsense.ai delivered a world class seminar focused on machine learning in the enterprise at AI World. This year, we have updated the content and have made deep learning a focal point of the seminar and the event.” said Eliot Weinman, Conference Chair and CEO, Trends Equity, Inc. AI World is the “must attend” event for enterprise executives and decision makers from Global 2000 organizations. The three-day conference and expo brings together the entire applied AI ecosystem, including innovative enterprises, industry thought leaders, startups, investors, developers, independent researchers and leading solution providers.

Prior to the event, deepsense.ai’s senior data scientist Michal Tadeusiak will conduct a webinar providing an overview of machine learning business applications to help participants discover possible use cases and define AI strategy. Those interested in getting a taste of what AI World is all about are invited to tune in on August 30th at 1 pm EST. More details and register here.

###

About deepsense.ai – AI World’s Strategic Machine Learning Partner:deepsense.io descripion

About AI World:
AI World Conference and Expo is focused on the business and technology of artificial intelligence in the enterprise. AI World 2017 will be held in Boston, MA on December 11-13. The three-day conference and expo is designed for business and technology executives who want to learn about the state-of-the-practice of AI in the enterprise. AI World has become is the “must attend” event for enterprise executives and decision makers from Global 2000 organizations and business leaders from across the entire artificial intelligence and machine learning ecosystem. AI World 2016 was the largest independent enterprise AI event ever produced, hosting more than 2,200 business executives, 110+ speakers and 65+ sponsors and exhibitors. To learn more, please visit www.aiworld.com.

Brand Manager at deepsense.ai

Workshop Manager at deepsense.ai

deepsense.io trademarks at boilerplate

Intel has invited deepsense.io to co-operate on the Intel Student Developer Program for AI, which was designed to help students learn the practical side of machine learning. The program builds on the success Intel has achieved in the US.

The program starts with Intel Machine Learning Seminar Powered by deepsense.io – a series of seminars geared to students from various technical faculties in Poland. The sessions include technical talks led by experienced data scientists from deepsense.io, a networking session and challenge for participants. Attendees will be taught the foundations of machine learning and deep learning, and see how Intel uses data science in its work. In the networking sessions, students will have a chance to talk personally with instructors and get to know each other in a convivial atmosphere.

“The Intel Student Developer Program for AI enables Intel to work with participating universities to bring exclusive access to newly optimized frameworks and technologies, machine learning training, and workshops for students, data scientists, and developers”, said Florian Piroth, EMEA Program Manager, Intel SSG Developer Relations.
“Data science is the sexiest job of the 21st century. We will show why by giving examples (both practical and cute) of data visualization and machine learning. We will also provide a short deep learning tutorial (with Keras in Python) so you can start as well”, added Piotr Migdal, Data Scientist and leading instructor at deepsense.io.

The seminars will be held at the top Polish technical universities:
May 22, 2017 – the Gdansk University of Technology (Politechnika Gdanska)
May 25, 2017 – the Warsaw University of Technology (Politechnika Warszawska)
May 26, 2017 – Wroclaw University of Science and Technology (Politechnika Wroclawska)
Seminars in Cracow, Warsaw, and Poznan will be announced soon.

Intel Technology Poland has been operating in Poland for 18 years now. During this time, the company has hired about 2,000 employees and built one of the biggest Intel research and development centers in Europe. Artificial Intelligence, Big Data, Embedded Systems, Cloud Computing, Compilers, and Non-volatile memory storage are just a few areas of expertise the company leverages in projects here.

“Be a part of an exciting new program for university students and engage with Intel around our work in machine learning, deep learning, and artificial intelligence”, encouraged Florian Piroth.

###

About deepsense.io:deepsense.io descripion

Brand Manager at deepsense.io

Workshop Manager at deepsense.io

deepsense.io trademarks at boilerplate

The deepsense.io team makes a significant contribution to the satellite imagery project hosted at Kaggle.com aimed at increasing the efficiency of the decision-making process for the UK’s defence and security.

The Defence Science and Technology Laboratory (Dstl), a UK government agency, recently hosted a competition to identify and categorize objects in satellite imagery. Dstl provided 1 km x 1 km satellite images and the challenging task was to detect different types of objects, such as buildings, vehicles, trees or roads located in various environments. Users had to find an algorithm or develop software that would help evaluate large and complex data sets in an innovative manner.

“Coming up with a model that can generalize well from a small training set of only 25 multispectral images was a really challenging task. Convolutional neural networks allowed us to create a reliable solution and take advantage of recent developments in the hot area of deep learning. We were excited to apply novel scientific solutions for practical problems and see their excellent performance,” said Robert Bogucki, Chief Science Officer at deepsense.io.

deepsense.io’s data science team took fourth place with their creative solution, beating over 400 other teams from all over the world. There were over 5,000 submissions on Kaggle, which translates into over £2.5m worth of research. The results can be used in a great number of projects requiring a quick and accurate analysis of large data sets.

“I am proud of our machine learning team, who proved their data science skills in the Kaggle competition once again. We took such a high place due to the fact that we have outstanding and talented people who use appropriate methodology and tools, such as Neptune, developed by us, initially created for our data science purposes, which has now been shared with the market. Thanks to such non-commercial projects we can develop our know-how and continuously improve working techniques used later in business challenges,” said Piotr Niedzwiedz, CTO at deepsense.io.

###

About deepsense.io:deepsense.io descripion

Read our blog: deepsense.ai/blog/

Try NeptuneGo! – an easily accessible public version of Neptune: go.neptune.deepsense.io

Brand Manager at deepsense.io

deepsense.io trademarks at boilerplate

deepsense.io is pleased to announce the results of its multi-month machine learning-based collaboration with the United Nations Office of Information and Communications Technology (ICT).

On November 3, 2016, the Company’s data scientist delivered a “TechNovation Brief – Machine Learning” seminar to United Nations’ officials and staff at the UN Headquarters in New York City. The presentation dealt with two case studies developed especially for ICT: one on global news monitoring and topic retrieval and the other on identifying propaganda disseminated via Twitter. These projects were used to illustrate machine learning’s potential uses in recognizing and categorizing requested information from large data inputs. During the news project’s categorization process, our scientists were happy to find that most of the topics were related to UN Sustainable Development Goals.

“In my view, the importance of this project goes well beyond the modeling results by providing an opportunity to explore how data science can impact programmatic needs on use cases related to UN mandates. Presenting these case studies and results at the UN’s first TechNovation briefing, exposes UN staff from different parts of the institution both to machine learning as a powerful tool and to the design thinking that will be useful in implementing this tool. We have greatly enjoyed this opportunity to work with deepsense.io.” Radia Funna, Head of Innovation, Office of Information & Communications Technology.

“Machine learning can be applied to data analysis and automatization, and it can improve global organizations’ and companies’ everyday processes. We were very happy to share our knowledge and experience with the Office of Information & Communications Technology at the United Nations,” added Tomasz Kulakowski, CEO at deepsense.io.

Monitoring and topic retrieval for global news
The project began with collecting United Nations-related news items (raw text) from around the world. The news items were collected through the GDELT project database which describes important news events and provides a reference to corresponding articles online. After downloading and pre-processing the news our team used the Latent Dirichlet algorithm to summarize the content of the news by a set of topics. The project’s primary focus was on unsupervised learning methods capable of inferring trending topics from a set of documents without additional human input. The results identified a number of topics related to issues such as African development, women’s health, current conflicts in the Middle East, nuclear weapons’ proliferation and climate change. Such analyses are helpful in monitoring the carrying out of the Sustainable Development Goals or getting insights into how a given topic is covered by the media.
“Machine learning seems particularly effective in two situations: systems that are unpredictable and constantly changing and systems that are so complex that we cannot fully describe them in one model. The UN is facing both of these: understanding how the world works is infinitely complex, and it changes all the time. That is why I believe machine learning can be of value to us.” Lambert Hogenhout, Chief Data Analytics, Innovation and Partnerships at United Nations

Detecting propaganda on Twitter
The second project dealt with monitoring Twitter posts for propaganda aimed at a particular target group. Processing Twitter data is challenging due to the enormous number of Tweets posted every second, their brevity and informal tone. To obtain data about a particular Twitter account, the deepsense.io team combined user-account statistics (number of Tweets and followers) with text analytics methods to process the content published by individual users. This data enabled the team to classify an account to a particular target group. The definition of a target group is arbitrary, for example, it might be a group of users supporting a particular idea/movement.

Inside “TechNovation Brief – Machine Learning” UN Seminar
The deepsense.io projects concluded with an internal UN “TechNovation Brief” seminar held on Nov. 3, 2016. During the meeting, officials from different UN agencies and departments had a chance to learn how powerful machine learning technologies could be used in their work. In addition to presenting the projects’ results, the seminar also included a discussion and a short demo on internal UN machine learning projects (NLP, Entity Recognition, Bots (Global Services Division)), as well as a general review of using machine learning predict domestic violence.
deepsense.io looks forward to continuing its cooperation with the United Nations’ Office of Information and Communications Technology in the future.

###

About Information and Communications Technology at the United Nations:
The Office of Information and Communications Technology is responsible for defining strategic direction for ICT to the Secretariat. It provides oversight of ICT programmes, budgets and decision-making to ensure alignment with the Secretariat’s overall ICT strategy. The team consists of more than 300 people operating global data centers and coordinating services provided by more than 4,000 people in UN offices worldwide. The Office is headquartered in New York and sits within the Department of Management.

About deepsense.io:deepsense.io descripion

Brand Manager at deepsense.io

deepsense.io trademarks at boilerplate