Just links
That's just link aggregator of everything I consider interesting, especially DL and topological condensed matter physics. @EvgeniyZh
Show moreData loading in progress...
Find out who reads your channel
This graph will show you who besides your subscribers reads your channel and learn about other sources of traffic.Posts | Views | Shares | Views dynamics |
01 DrEureka: Language Model Guided Sim-To-Real Transfer https://eureka-research.github.io/dr-eureka/ | 325 | 4 | Loading... |
02 Topological holography for fermions https://arxiv.org/abs/2404.19004 | 498 | 2 | Loading... |
03 Time-Reversal Anomalies of QCD3 and QED3 https://arxiv.org/abs/2404.17233 | 591 | 3 | Loading... |
04 Training on Mistakes Improves Problem Solving Performance https://gytis.co/2024/04/25/Training-on-Mistakes-Improves-Problem-Solving-Performance.html | 1 238 | 22 | Loading... |
05 https://www.lesswrong.com/posts/jGuXSZgv6qfdhMCuJ/refusal-in-llms-is-mediated-by-a-single-direction | 1 292 | 14 | Loading... |
06 В Абу-Даби прошла первая в мире гонка болидов-беспилотников.
Самое интересное происходило на квалификации, когда машины делали тестовые заезды, чтобы «познакомиться» с трассой. Машины вылетали с трасс, разворачивались, влетали в отбойники. Одним словом делали то, за что мы любим беспилотники. | 1 017 | 17 | Loading... |
07 https://twitter.com/percyliang/status/1783724077972361647 | 940 | 4 | Loading... |
08 Offline Actor-Critic Reinforcement Learning Scales to Large Models
В основном ресерч в сфере рл происходит на маленьких моделях пушто
- и небольшие модели способны решить задачи при грамотном обучении и грамотной архитектуре (эмпирический факт), в основном это млп с релу/лики_релу и леернормой и все
- есть много проблем в сетапе рл, которые надо решать до того, как задумываешься о модельке и ее размерах
- если увеличивать размер модели, то это доп проблемы, потому что увеличиваются риски для нестабильности, вырожденности и всего такого
Но дипмаинд решил разнести и эту парадигму и отскейлить рл модели до больших размеров
Как оказывается, актор-критик в совокупности с perceiver'ом, который здесь может обрабатывать разные стейты для разных роботов (или симулякров роботов) + постепенно отходить от бихевиор клонинга, и выбивать высокий скор как на средах, где данные собраны хорошо, так и плохо!!
И это все на 132 тасках с непрерывными действиями🥸
👀LINK
#rl #offlinerl #multitask #behaviorcloning #largemodels #scalinglaws | 822 | 19 | Loading... |
09 Anyonic Topological Order in Twisted Equivariant Differential (TED) K-Theory arxiv.org/abs/2206.13563 | 781 | 3 | Loading... |
10 https://sites.google.com/view/ph11fall2019/previous-hurdles | 883 | 3 | Loading... |
11 Preference Fine-Tuning of LLMs Should Leverage Suboptimal, On-Policy Data https://arxiv.org/abs/2404.14367 | 1 021 | 9 | Loading... |
12 Из других свежих интересностей, HF опубликовал свою открытую реимплементацию Gato (https://t.me/gonzo_ML/966) под названием Jack of All Trades (JAT).
Пост: https://huggingface.co/blog/jat
Статья: https://arxiv.org/abs/2402.09844
Код: https://github.com/huggingface/jat
Модель: https://huggingface.co/jat-project/jat
Датасет: https://huggingface.co/datasets/jat-project/jat-dataset | 875 | 12 | Loading... |
13 Lattice Surgery for Dummies https://arxiv.org/abs/2404.13202 | 1 013 | 5 | Loading... |
14 COCONut: Modernizing COCO Segmentation arxiv.org/abs/2404.08639 | 1 363 | 9 | Loading... |
15 Demonstration of logical qubits and repeated error correction with better-than-physical error rates https://arxiv.org/abs/2404.02280 | 1 430 | 5 | Loading... |
16 Scaling Instructable Agents Across Many Simulated Worlds https://arxiv.org/abs/2404.10179 | 1 467 | 16 | Loading... |
17 Probing the 3D Awareness of Visual Foundation Models arxiv.org/abs/2404.08636 | 1 834 | 9 | Loading... |
18 #prog #game #abnormalprogramming
Verilog to Factorio blueprints compiler
https://github.com/Redcrafter/verilog2factorio
(thanks @GabrielFallen) | 1 728 | 10 | Loading... |
19 Direct transition from a fractional quantum anomalous Hall state to a smectic state with the same Hall conductance https://arxiv.org/abs/2404.06745 | 1 338 | 1 | Loading... |
20 Certifying almost all quantum states with few single-qubit measurements https://arxiv.org/abs/2404.07281 | 1 363 | 1 | Loading... |
21 Wu's Method can Boost Symbolic AI to Rival Silver Medalists and AlphaGeometry to Outperform Gold Medalists at IMO Geometry
китайцы сказали что ваши альфа геометри хуйня, у нас есть method Wu дома, прикрутили к нему cетку для выбора лучших бимов из alpha geometry и стало лучше чем было
paper | 1 191 | 8 | Loading... |
22 Higher Landau-Level Analogues and Signatures of Non-Abelian States in Twisted Bilayer MoTe2 https://arxiv.org/abs/2404.05697 | 1 319 | 1 | Loading... |
23 LLM2Vec: Large Language Models Are Secretly Powerful Text Encoders https://arxiv.org/abs/2404.05961 | 1 589 | 37 | Loading... |
24 Google Axion Processors – Arm-based CPUs designed for the data center (Score: 151+ in 5 hours)
Link: https://readhacker.news/s/65Ehb
Comments: https://readhacker.news/c/65Ehb | 1 361 | 1 | Loading... |
25 Decoupling Static and Hierarchical Motion Perception for Referring Video Segmentation arxiv.org/abs/2404.03645 | 1 442 | 6 | Loading... |
26 JRDB-PanoTrack: An Open-world Panoptic Segmentation and Tracking Robotic Dataset in Crowded Human Environments arxiv.org/abs/2404.01686 | 1 642 | 8 | Loading... |
27 Non-Abelian braiding of Fibonacci anyons with a superconducting processor arxiv.org/abs/2404.00091 | 1 476 | 3 | Loading... |
28 Submitted on 4 Apr 2024] Training LLMs over Neurally Compressed Text https://arxiv.org/abs/2404.03626 | 1 512 | 19 | Loading... |
Topological holography is a conjectured correspondence between the symmetry charges and defects of a $D$-dimensional system with the anyons in a $(D+1)$-dimensional topological order: the symmetry...
Anomalies of global symmetry provide powerful tool to constrain the dynamics of quantum systems, such as anomaly matching in the renormalization group flow and obstruction to symmetric mass...
This work was produced as part of Neel Nanda's stream in the ML Alignment & Theory Scholars Program - Winter 2023-24 Cohort, with co-supervision from…
HELM Lite v1.2.0 is out! Datasets: NarrativeQA, NaturalQA, OpenbookQA, MMLU, MATH, GSM8K, LegalBench, MedQA, WMT14 Results (we still need to add Claude 3, which requires more prompt finagling):
https://t.co/7DMWXmpey3While the classification of non-interacting crystalline topological insulator phases by equivariant K-theory has become widely accepted, its generalization to anyonic interacting phases -- hence...
Hurdles are questions that do not necessarily have correct answers. A successful solution approaches the problem thoughtfully and creatively, using modelling to address the problem. Several assumptions may need to be made over the course of your answer. Each calculation does not need to be