eminorhan Goto Github PK

adept

adept intuitive physics benchmark data & tools

alternative-imagefolders

Two useful variations on ImageFolder

annealed-attention

an experimental annealed self-attention layer

baby-vision

Self-supervised learning through the eyes of a child

babylm

Experiments for the BabyLM challenge

babylm-evals

bold-5000

fMRI signals from visual cortical areas are too noisy to distinguish between drastically different image recognition models

cluster-support

cogs-pretrained-lms

experiments with pretrained LMs on the COGS benchmark

complex-neurons

complex neurons, cell types, and all that jazz

dino

DINO

dit

DiT clone

faster-dit

An even faster DiT clone

gutenberg

human-ssl

How much "human-like" visual experience do current self-supervised learning algorithms need to achieve human-level object recognition?

humanlike-vits

ViT models pretrained with up to ~5k hours of human-like video data

hvm-1

Video models trained on ~5k hours of human-like video data

igpt

Image GPT

igpt-memory

Can deep learning match the efficiency of human visual long-term memory for object details?

igpt-ood

Probing the ood behavior of OpenAI iGPT models

inevitable-probability

Efficient probabilistic inference in generic neural networks trained with non-probabilistic feedback

llm-memory

Memory experiments with LLMs

mae

Masked Autoencoders for image-based SSL

mae_st

Spatiotemporal Masked Autoencoders for video-based SSL

mixture-of-experts

Mixture of experts layers for Keras

models-for-guy

pretrained models for Guy's relation learning project

mugs

Mugs

nano-gpt

a clone of Andrej Karpathy's nano-gpt implementation

nonnormal-init

Sequential non-normal initializers for recurrent neural networks

numerical-fisher

Numerical optimization of Fisher information for a parametrized population code