15 Outstanding Research Papers from NeurIPS 2025
▪️ Faster R-CNN
▪️ Artificial Hivemind: The Open-Ended Homogeneity of LMs (and Beyond)
▪️ Optimal Mistake Bounds for Transductive Online Learning
▪️ Gated Attention for LLMs
▪️ Superposition Yields Robust Neural Scaling
▪️ Why Diffusion Models Don’t Memorize
▪️ Does RL Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?
▪️ 1000 Layer Networks for Self-Supervised RL
▪️ Google's Titans + MIRAS
▪️ Generative Data Augmentation via Diffusion Distillation, Adversarial Alignment, and Importance Reweighting
▪️ Slow Transition to Low-Dimensional Chaos in Heavy-Tailed RNNs
▪️ Evaluating multiple models using labeled and unlabeled data
▪️ Riemannian Consistency Model
▪️ BioReason
▪️ NFL-BA: Near-Field Light Bundle Adjustment for SLAM in Dynamic Lighting
Save the list, and check this out for the paper links and more info: https://t.co/r1TqNe4qQp
Last post in the @TDataScience series on expressive power of #graphneuralnet I suggest using metric geometry to address a broader setting than graph isomorphism. Next post: do we need deep #GNN on graphs?
@benhamner I suppose by non-English speaking SE you mean non-native english speaker? Most of us learn english at some stage when getting into SE as documentation and code tend to enforce this. Myself, I used to struggle with some words like "fetch" to understand what the functions does
Donald Knuth is a computer scientist who understands that words matter. His works include a philosophy of literate programming, the typography program TeX and “The Art of Computer Programming,” which he began writing in 1962 and still hopes to complete. https://t.co/RreN9xWsMb
Here's my conversation with David Silver, who leads the reinforcement learning group at @DeepMind. He lead the AlphaGo & AlphaZero research effort, which to me, taken together, is the greatest accomplishment in the history of artificial intelligence: https://t.co/JoAJ9LXxIk
Your first project will be horrible.
Your first web app will suck.
Your first Client will be hell.
Your first project will lose money.
Your first tweet will get 0 likes.
But you can't make your 50th without making your first.
So get over it & DO IT.
#100DaysOfCode#100DaysOfCode
I recommend not only social distancing but social media distancing. Read books. Take classes. Work on a difficult project. Start a personal revolution. I tune in 2-3 times a day, read quickly, smile, and go back to work. Just enough for the good vibes. Love you all. Stay strong!