Replay

Gwern — Deep Learning

by Gwern Branwen

Neural network research, scaling, and architectures.

12 posts

Loading...

Delivery order

Each email contains one post, starting with #1

1

The Kelly Coin-Flipping Game: Exact Solutions

38 min

We can approximate it with our pre-existing value function for a known stopping time/edge/max wealth, sampling from the posterior; for example, we might draw 1000 values from 𝒩(300,25), the Pareto,...

Deep Learning
2

Free-Play Periods for RL Agents

8 min

Proposal for incentivizing meta-learning of exploration in deep reinforcement learning: domain randomization with reward-shaping, where there is a fixed-length ‘play time’ with no rewards/losses at...

Deep Learning
3

Novelty Nets: Classifier Anti-Guidance

10 min

Generative modeling proposal for increasing diversity of samples by a helper NN memorizing past samples and ‘repelling’ new samples away from old ones.

Deep Learning
4

Number Search Engine via NN Embeddings

5 min

Proposal to create a ‘search engine’ like OEIS but for individual numbers, allowing fuzzy lookups, by training a neural net embedding on the scientific & mathematic literature’s corpus of...

Deep Learning
5

Research Ideas

36 min

Choose-Your-Own-Adventure generative fiction for efficiency/editing (2021-06-06) CYOA generative fiction Try directly optimizing reward generation (2019-12-16): backpropping reward...

Deep Learning
6

Absolute Unit NNs: Regression-Based MLPs for Everything

7 min

One might wonder: is an AUNN truly the very simplest possible NN architecture? Maybe not. We still have the index input making it more complex.

Deep Learning
7

LLM Daydreaming

10 min

Proposal & discussion of how default mode networks for LLMs are an example of missing capabilities for search and novelty in contemporary AI systems.

Deep Learning
8

‘end-to-end’ directory

1 min

https://arxiv.org/abs/2106.10316#deepmind : “Proper Value Equivalence” , Christopher Grimm, André Barreto, Gregory Farquhar , David Silver , Satinder Singh link-bibliography https://arxiv.

Deep Learning
9

‘NN sparsity’ directory

2 min

https://arxiv.org/abs/2510.15103#facebook : “Continual Learning via Sparse Memory Finetuning” , Jessy Lin, Luke Zettlemoyer , Gargi Ghosh , Wen-Tau Yih, Aram Markosyan, Vincent-Pierre Berges, Barlas...

Deep Learning
10

‘AI scaling’ directory

32 min

https://www.sciencedirect.com/science/article/pii/S016028962500025X : “Psychometrically Derived 60-Question Benchmarks: Substantial Efficiencies and the Possibility of Human-AI Comparisons” , Gilles...

Deep Learning
11

‘MLP NN’ directory

6 min

https://arxiv.org/abs/2503.24187 : “NeuRa L a T e X : A Machine Learning Library Written in Pure L a T e X ” , James A. D. Gardner, Will Rowan, William A. P. Smith link-bibliography https://arxiv.

Deep Learning
12

‘self-attention’ directory

7 min

https://arxiv.org/abs/2507.02754#facebook : “Fast and Simplex: 2-Simplicial Attention in Triton” , Aurko Roy, Timothy Chou, Sai Surya Duvvuri , Sijia Chen, Jiecao Yu, Xiaodong Wang , Manzil Zaheer,...

Deep Learning