Mikhail Burtsev Email & Phone Number

London, England, Gb

Belmont, Ca, Us

DeepPavlov (DP) is an open-source framework to develop and deploy conversational assistants in production.Who are users of DeepPavlov?* Developers of multi-skill conversational agents and stand alone skills. Use DP to build and deploy production ready systems. Reuse and finetune existing components, skill templates and reference conversational agents. Contribute new skills and industry related features to the DP.* Researchers working in the area of Conversational AI. Use DP to experiment with new ideas for Conversational AI. Reuse referent agents to save time on implementation of infrastructure for dialogue task, and reuse basic models to create new architectures. Contribute new models for conversational functions.

Москва, Москва, Ru

Prior to foundation of the AIRI designed core organizational principles, after establishment of the institute worked on a mission and key goals. Set up internal science board. Designed and implemented research procedures for internal scientific projects including proposal structure, submission and evaluation by internal science board. Interviewing key employees out of total 90 research staff. Review of more than 25 research project proposals. Development of AIRI’s basic research strategy which was successfully approved by external science advisory board.

Dolgoprudniy, Moscow Region, Ru

Set up and run lab currently including of 3 postdocs, 10 PhD students, 5 MS students, 3 BS students, 5 software engineers, 4 administrative staff. Started development of open-source conversational AI framework DeepPavlov. DeepPavlov has about 360 000 pip installs over 4 years with current monthly download rate of 15 000. DeepPavlov was recognized as one of the winners of Google’s “Powered by TF challenge” and selected as organization included in the Google Summer of Code 2021 program. Lab was selected as NVIDA and Facebook academic partner in 2016-2017. Supervised team of PhD and MS students who participated in Amazon Alexa 3 and 4 Challenges (supported by $500K funding from Amazon). In 2021 lab won the Read//able UpGreat Prize (about $1 mln.) for the human level teacher like proof-reading and commenting of high school essays.

Brussels, Be

Review of three research projects for European Innovation Council and Small and Medium-sized Enterprises Executive Agency (EISMEA).

ConvAI2 NeurIPS competition that aims to further the state-of-the-art in open-domain chatbots. Some key takeaways from the competition are:(i) pretrained Transformer variants are currently the best performing models on this task,(ii) but to improve performance on multi-turn conversations with humans, future systems must go beyond single word metrics like perplexity to measure the performance across sequences of utterances (conversations)--in terms of repetition, consistency and balance of dialogue acts (eg how many questions asked vs. answered).

The first Conversational Intelligence Challenge ( convai.io/2017 ) was conducted over 2017 with finals at NIPS conference. The challenge aimed at evaluating the state-of-the-art in non-goal-driven dialogue systems (chatbots) and collecting a large dataset of human-to-machine and human-to-human conversations manually labelled for quality. We established a task for formal human evaluation of chatbots that allows to test capabilities of chatbot in topic-oriented dialogue. Instead of traditional chit-chat, participating systems and humans were given a task to discuss a short text. Ten dialogue systems participated in the competition. The majority of them combined multiple conversational models such as question answering and chit-chat systems to make conversations more natural. The evaluation of chatbots was performed by human assessors. More than 1,500 volunteers were attracted and over 4,000 dialogues were collected during competition. Final score of the dialogue quality for the best bot was 2.746 compared to 3.8 for human. This demonstrates that current technology allows supporting dialogue on a given topic but with quality significantly lower than that of human. To close this gap we plan to continue the experiments by organising the next conversational intelligence competition. This future work will benefit from the data we collected and dialogue systems that we made available after the competition.

The school-hackathon http://RL.DeepHack.me had the task of improving performance of DNN RL algorithms on those Atari games where deep learning approaches still demonstrate subhuman performance. Leading scientists working in the field of reinforcement learning including Andrew Barto and Doina Precup gave talks to eight teams totaling 47 hackers participated in the event.Video of the final match - https://youtu.be/Hq2dVKviowILeaderboard http://deeprl.tilda.ws/page95443.htmlDeepHack.RL is distinguished from similar events which set practical goals by its emphasis on scientifically useful results applicable to AI algorithm development. Side by side with the coding challenge, a science school offered lectures by leading specialists in the field of machine learning, which were attended by over 300 people.

Department of Neuroscience, Kurchatov NBIC Centre, Moscow, RussiaDesigned and launched cell culture laboratory equipped with multi-electrode system for recording and stimulating activity of neurons in vitro. Supervised experiments to study learning in networks developed from primary hippocampal cultures of mice neurons. Simulation of biologically plausible neural networks to model learning in vitro. Proposed novel neuroevolutionary methods.

Winter deep learning school-hackathon Deephack Q&A ( http://qa.deephack.me/ , talks areavailable at https://goo.gl/nQ49MS ). People from DeepMind (Phil Blunsom), OpenAI, FAIR, Google, NYU, and Stanford University gave talks to more than 50 participants who have been hacking the Kagglechallenge on the 8th grade test questions answering.

Vice Chair of ICT Panel for the Horizon 2020 Future and Emerging Technologies (FET) projects.Remote work consisting of cross-reading FET Open proposals and Individual Evaluation Reports (IERs) resulting from the remote evaluation. Meeting in Brussels for the final panel review. Acting as Rapporteur for a set of about 10 FET OPEN RIA proposals and as Crossreader in the panel discussion for another set of about 30 proposals.

Host prof. Patrick Bateson

Cambridge, Gb

Host prof. Patrick Bateson

Doctor Of Philosophy (Ph.D.), Computer Science

Complex Systems

Master Of Science (Msc), Microelectronics