YK Community Feed

#NeurIPS2022 What are ideal representations for self-sup. learning (SSL)? 🤓We give simple optimality conditions and use them to improve/understand/derive SSL methods! 🔥outperform baselines on ImageNet arxiv.org/abs/2011.10566 w. @tatsu_hashimoto @StefanoErmon @percyliang 🧵

#NeurIPS2022 What are ideal representations for self-sup. learning (SSL)? 🤓We give simple optimality conditions and use them to improve/understand/derive SSL methods! 🔥outperform baselines on ImageNet arxiv.org/abs/2011.10566 w. @tatsu_hashimoto @StefanoErmon @percyliang 🧵

This may revolutionize data science: we introduce TabPFN, a new tabular data classification method that takes 1 second & yields SOTA performance (better than hyperparameter-optimized gradient boosting in 1h). Current limits: up to 1k data points, 100 features, 10 classes. 🧵1/6

Happy to finally share our work! Joint collaboration with Louay Hazami and Ragavan Thurairatnam. Efficient-VDVAE obtain SOTA log-likelihood on 7 commonly used image datasets. We make an argument against 5-bit benchmarks and lay some ground work for downstream tasks with HVAEs.

FIGARO: Generating Symbolic Music with Fine-Grained Artistic Control abs: arxiv.org/abs/2201.10936

I’m excited to present our paper CM3: A Causal Masked Multimodal Model of the Internet where we train a model that can do zero-shot unconditional/conditional image generation (PixelCNN/DALL-E), image-infilling/captioning, entity linking/disambig, summarization all with prompting!

Are all bits useful? How can we simplify images, and how do “simpler” images affect neural network training?🤔 with @sarahookr @savvyRL Tonio Ball Paper: arxiv.org/abs/2201.05610 Code: tinyurl.com/simple-bits 🧵

Excited to share my newest paper, "Neural Language Models are Effective Plagiarists" with @EdwardRaffML. We took a dataset of CS 101 assignments and asked "can a language model do a good job solving these with minimal human intervention or knowledge?" arxiv.org/abs/2201.07406

We had the privilege of presenting our Sim2Real Docs project at @NeurIPSConf 2021 DCAI. TLDR: Python library for synthetic images of documents in natural scenes using @Blender. It's open source: github.com/fidelity/sim2real… It's @nmaddikunta21's first paper: arxiv.org/abs/2112.09220

We had the privilege of presenting our Sim2Real Docs project at @NeurIPSConf 2021 DCAI. TLDR: Python library for synthetic images of documents in natural scenes using @Blender. It's open source: github.com/fidelity/sim2real… It's @nmaddikunta21's first paper: arxiv.org/abs/2112.09220

Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasets abs: arxiv.org/abs/2201.02177 NNs learn through a process of "grokking" a pattern in the data, improving generalization performance from random chance level to perfect generalization

When minimizing objectives, randomly initializing + optimizing can fail to find different solutions. We give a branching optimization method for this using Lyapunov exponents..1/11 arxiv.org/abs/2112.14570 @AAMAS2022 With @PaulVicol @jparkerholder @TalKachman @Luke_Metz @j_foerst

ArXiv arxiv.org/abs/2112.15594: Transformers pre-trained on text and fine-tuned on code solve math problems by program synthesis (questions as Codex prompts). Solves and grades problems from MIT's math courses and a math benchmark perfectly. Generates new math questions.