ML Review 🇺🇦

✨🎓🏆 Machine Learning Digest №11 🏆🎓✨ Highlighted Research Papers, Projects, Articles, and Videos. mlreview.com/digest/TensorF…

Elegant idea @AndrzOdrz shows standard elementary functions can be built from just EML(x,y)=exp(x)-ln(y) and 1 Boolean logic - NAND Elementary math - EML. @PMocz built a two-button calc - github.com/pmocz/two_butt… Useful for symbolic-regression grammar arxiv.org/abs/2603.21852

5/ Adversarial Attacks 🔹LLM-Inherited: jailbreaks, prompt injections, data extraction 🔹Agent-Specific: knowledge poisoning, untrusted tool call, backdoor insertion, environment injection 🔓 Jailbreak Methods: Universal suffix gradient search, AutoDAN, and Crescendo

4/ RAG → HippoRAG (RAG + Knowledge Graph + PageRank) by @bernaaaljg Outperforms SoTA multi-hop QA by up to 20% While being 10-30x faster and 6-13x cheaper. 📄 Paper: arxiv.org/abs/2405.14831 🐙 Code: github.com/OSU-NLP-Group/…

1/ "Language Agents: Foundations, Prospects, and Risks" #emnlp2024 Tutorial by @ysu_nlp @Diyi_Yang @ShunyuYao12 @taoyds 📑 Slides: tinyurl.com/language-agent… 🎥 Video: tinyurl.com/agent-tutorial…

"The Little Book of Deep Learning" 140 pp📖 Book by @francoisfleuret 🫡 fleuret.org/public/lbdl.pdf

@unsorsodicorda @solvay_1927 GPT4All, Dolly, Flan-Alpaca, and ChatRWKV as "no commercial use" instruct-tuned LLMs And OpenChatKit as the only true Apache 2.0. This time I'd also include FLAN-T5

@ml_review @solvay_1927 What where the models (and the repo) you mentioned in the tweet you deleted?

@solvay_1927 Plausible, if so I was misled by Alpaca release: "the instruction data is based on OpenAI’s text-davinci-003, whose terms of use prohibit developing models that compete with OpenAI" Sadly, all of the above models still suffer from LLaMa or Alpaca restrictive license

@ml_review OAI ToS only applies I think to the person directly querying OAI for generations. AI generations however can't be copyrighted and so a third party which has not used OAI for making generations to train models is not violating the ToS.

@dezhou @DivGarg9 @markchen90 @giffmana @adityagrover_ @barret_zoph @drew_jaegle @aidangomezzz @ch402 Hope not from the movie, there is only One True "Transformers"

@ml_review @DivGarg9 @markchen90 @giffmana @adityagrover_ @barret_zoph @drew_jaegle @aidangomezzz @ch402 Is the name `Transformers` borrowed from the movie "Transformers"?

"Stanford CS25 - Transformers United" Seminar by @DivGarg9 @markchen90 @giffmana @adityagrover_ @barret_zoph @drew_jaegle @aidangomezzz @ch402 Geoffry Hinton Covers Transformers in Language, Vision, Decision-Making, Audio youtube.com/playlist?list=…

Pen and Paper Exercises in Machine Learning by Michael U. Gutmann arxiv.org/abs/2206.13446 kudos to the author for putting these online -- surely will be a valuable resource for educators of all stripes (but especially for TAs and those who teach privately).

🔮Nixtla - Forecasting Pipeline as a Service By @fede_gr – 25% better accuracy than Amazon Forecast – 20% more accurate than fbprophet – 4x faster than Amazon Forecast & less expensive github.com/Nixtla/nixtla

Close The Sky Over Ukraine

Bayesian Modeling and Computation in Python Online Book by @aloctavodia @canyon289 @junpenglao Powered by Jupyter Book bayesiancomputationbook.com/welcome.html

glum - fast, maintainable, Python-first library for fitting GLMs with an extensive feature set. By @tbenthompson -L1/L2/Tichonov –Normal,Poisson,Tweedie etc –Box/linear inequality constraints,sample weights, offsets –scikit-learn compatible github.com/Quantco/glum

"Deep Learning for Molecules and Materials. 2021" Online book by @andrewwhite01 dmol.pub

The Roadmap of Mathematics for Machine Learning By @TivadarDanka tivadardanka.com/blog/roadmap-o…

An Elementary Introduction to Information Geometry Survey by @FrnkNlsn [2020, 56 pp] - Basics of Differential Geometry - Information Manifolds - Applications abs arxiv.org/abs/1808.08271 pdf arxiv.org/pdf/1808.08271…

Excited to announce Dodrio, an interactive visualization tool designed to help NLP researchers and practitioners analyze and compare attention weights in transformer-based models with linguistic knowledge! Try Dodrio in your browser: poloclub.github.io/dodrio/ #ACL2021NLP #NLProc