gkirgizov / hetarch_stm32f4

HETARCH project: runtime for stm32f429i-discovery target

👋 Hi, Stranger 👽

I'm Gregory, AI Research Engineer

📃 My CV is here

I'm working on hybrid intelligence methods and developing on methods with relevance beyond current AI wave. My mission is to make the inanimate world responsive to human needs through ambient intelligent technology. Technology must be alive and communicable. If this resonates with you, I'd like to chat. Please 💬 drop me a message. I'm always looking for fellow futurists and visionaries.

I worked at JetBrains on domain language design and incremental type checkers (project CodeRules). Currently I work on AutoML & evolutionary graph optimization methods at AI Institute in Saint-Petersburg @aimclub.

---

My Research 📜 & Development 🔭

My ongoing research includes:

• (paper WIP) GOLEM framework — R&D on the algorithm and architecture of the framework for graph optimization. Core points: universality of the approach and adaptability of the algorithm. Currently in work and use by several internal teams at ITMO AI Institute (@aimclub).
• (paper WIP) Evolution + Reinforcement Learning — a research attempt to unify two methods from different fields of AI. Core metaphor is that Evolution is not a blind principle, but an adaptive agent in genotypic space. Then RL provides adaptive local search, while Evolutionary algorithm controls global search in complex spaces. Win-win.
• (paper submitted) Rostok for robots — collaboration on automated robotic design using GOLEM as optimization core.
• (experiments WIP) Distributed Intelligence Environment — is an Artificial Life project aimed at reproducing emergence of distributed intelligence under environmental pressures using neural cellular automata based on Evolution, RL & Active Inference.

What I would like to collaborate on? 🚀

If your projects have anything in common with this, then maybe we can discuss something interesting!

• Code Generation — I believe that current AI state has under-researched potential for combinations for formal & LLM methods. We have very strong formal methods and code verifiers. On the other side, we have rich stochastic language models. I'd like to check the intersection of these.
• Smart Contracts — especially with connections to AI/ML. I view smart contracts as cooperating agents in a distributed environment. From such perspective you can think about few curious applications of AI.
• Reinforcement Learning — would like to work on practical RL applications, as I'm quite excited about this field.
• Active Inference — it's like physics of intelligent agents based on Bayesian principles, and it's like RL on steroids. Something to be heard of a lot in a coming decade.

Past publications & conferences ✨

• Publication: Kirgizov G.V., Kirilenko I.A., «Heterogeneous Architectures Programming Library», ISP RAS Proceedings, Vol. 30, iss. 4, pp. 45-62 (doi>10.15514/ISPRAS-2018-30(4)-3).
• Meetup talk: “Coderules, a new typechecking engine”, at MPS Users Conference in Amsterdam, Oct 2019.
• Publication: S. Lazareva, G. Kirgizov, and R. Ragimov. «Smart face control: machine learning algorithms for efficient SSD caching» Proceedings of the 13th Central & Eastern European Software Engineering Conference in Russia. ACM, 2017 (doi>10.1145/3166094.3166109).