Emin Orhan

Advanced
@eminorhan

baby-vision. Self-supervised learning through the eyes of a child

145

mixture-of-experts. Mixture of experts layers for Keras

93

silicon-menagerie. Menagerie of models trained on SAYCam (and more)

26

recurrent-memory. When do neural networks learn sequential vs. persistent solutions in short-term memory tasks?

16

resnext-wsl. Robustness properties of Facebook's ResNeXt WSL models

15

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

14

ood-benchmarks. Out-of-distribution generalization benchmarks for image recognition models

14

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

11

robust-vision. Improving the robustness of ImageNet classifiers against natural and adversarial perturbations

10

llm-memory. Memory experiments with LLMs

10

video-models. Menagerie of video models trained on various video datasets

10

vqgan-gpt. GPT on VQGAN

7

hvm-1. Video models trained on ~5k hours of human-like video data

6

dino. DINO

4

gutenberg. Python

4

neural-pile-rodent. spiking neural activity data from rodents

4

nonnormal-init. Sequential non-normal initializers for recurrent neural networks

3

third-order. an experimental third-order attention model

3

transformer-metaseq2seq. Meta-seq2seq with generic transformers

2

igpt. Image GPT

2

parsing-transformers. Compositional generalization in semantic parsing with pretrained transformers

2

mae_st. Spatiotemporal Masked Autoencoders for video-based SSL

2

simsiam. A SimSiam implementation with a pure MLP encoder

2

simple-cache. A simple cache model for image recognition

2

frontier-accelerate. llm training workbook for frontier

1

neural-pile-rodent-old. spiking neural activity data from rodents

1

optimized-mae. An optimized implementation of masked autoencoders (MAEs)

1

neural-pile-primate-old. spiking neural activity data from primates

1

vqgan-gpt-video. GPT on VQGAN for video

1

bold-5000. fMRI signals from visual cortical areas are too noisy to distinguish between drastically different image recognition models

1

relational-recurrent. Attempted reproduction of the relational LSTM model.

1

neural-pile-primate. spiking neural activity data from primates

1
32
Apply