Data Science by ODS.ai 🦜

2022-11-17 09:49:58 Seeing Beyond the Brain: Conditional Diffusion Model with Sparse Masked Modeling for Vision Decoding

TLDR: Scientists kinda learned how to read thoughts. Paper on the reconstruction of the visual stimuli based on fMRI readings.

Website: https://mind-vis.github.io
Github: https://github.com/zjc062/mind-vis

#fMRI #visualstimulireconstruction #mindreading #dl Open / Comment

2022-11-15 13:25:16 #events : ML-тренировка
Когда: 17 (четверг) ноября 2022, 19:00 - 21:30 (сбор с 18:00)
Место: офис Яндекса (Москва, улица Льва Толстого, 16) + онлайн
Язык - русский

В этот раз нас ждёт 3 доклада:
- призер только что завершившегося Yandex ML Cup,
- 2ое место хакатона AgroCode Hack по анализу спутниковых снимков для виноградников
- организатор ML соревнований в информационной безопасности

Подробная программа по ссылке ниже
Будем рады видеть всех очно и онлайн ;)
Регистрация обязательна Open / Comment

2022-11-15 13:25:16 (RU lang) Open / Comment

2022-11-13 12:29:07 Reinforcement learning course from MIPT.

The course consists of:
- Theoretical and practical material for beginners and advanced users
- Classical approaches based on utility functions and strategy gradient, as well as modern trends in improving the efficiency of the study of the environment, interaction with planning, using memory and hierarchical approaches.
- The best of David Silver's lectures, Sutton and Barto's book, and OpenAI, DeepMind works and articles from 2019-2022.

Materials:
- PDF slides and video lectures on each topic, Colab master classes and video lectures in Russian.

Course: https://clck.ru/32a3c9

If you are interested in an internship at MIPT in the areas of Reinforcement Learning, Computer Vision, Robotics or Self Driving Cars, you can apply here: https://cogmodel.mipt.ru/internship Open / Comment

2022-10-28 11:05:16 Amos: An Adam-style Optimizer with Adaptive Weight Decay towards Model-Oriented Scale

Amos is a new optimizer that we propose to pre-train large language models. It is more efficient and converges faster than AdamW: ≤ 51% memory for slot variables, and better valid loss within ≤ 70% training time!Amos is a new optimizer that we propose to pre-train large language models. It is more efficient and converges faster than AdamW: ≤ 51% memory for slot variables, and better valid loss within ≤ 70% training time!

ArXiV: https://arxiv.org/abs/2210.11693

#NLU #NLP #optimizer Open / Comment

2022-10-24 08:57:57 Just in case you didn’t know what to wear for Halloween party. Open / Comment

2022-10-23 12:51:45 State of AI Report 2022

TLDR: We are moving forward and effective international collaboration is the key to progress.

Major Themes:

* New independent research labs are rapidly open sourcing the closed source output of major labs
* Safety is gaining awareness among major AI research entities
* The China-US AI research gap has continued to widen
* AI-driven scientific research continues to lead to breakthroughs

Website: https://www.stateof.ai

#report #stateofai #AI Open / Comment

2022-10-19 18:00:30 CORL: Offline Reinforcement Learning Library

Offline RL is a fresh paradigm that makes RL similar to the supervised learning, thus making it better applicable to the real-world problems. There is a whole bunch of recently developed Offline RL aglorithms, however, there was nots many of reliable open-sourced implementations for such algorithms

Our friends from @tinkoffai do some research in this direction and they recently open-sourced their internal offline RL library.

The main features are:
- Single-file implementations
- SOTA algorithms (Decision Transformer, AWAC, BC, CQL, IQL, TD3+BC, SAC-N, EDAC)
- Benchmarked on widely used D4RL datasets (results match performances reported in the original papers)
- Wandb logs for all of the experiments

Hope you will like it and the whole new world of Offline RL!

Github: https://github.com/tinkoff-ai/CORL

#tinkoff #RL #offline_lib Open / Comment

2022-10-18 12:20:15 Sequoia Gen AI Market Map

#generativeai #sequoia #infographics Open / Comment

2022-10-13 11:43:08 Out of One, Many: Using Language Models to Simulate Human Samples

TLDR: GPT-3 has unexpected application — modelling of socialogical studies. Average responses of a certain groups can be to some algorithmical accuracy predicted by in silico modelling.

What this means: sociologists won’t have to conduct costly live researches and will be able to run experiments in simulations. Marketers and politicians are getting their hands on cheap solution for modelling their slogans and value propositions. This enables people to check more hypothesis faster and to manipulate society with more efficiency.

ArXiV: https://arxiv.org/abs/2209.06899

#gpt3 #psychohistory #nlu #sociology Open / Comment