苏剑林(Jianlin Su)

Papers.Cool. I built a website for browsing research papers like a social feed.

Keras-DDPM. I built a simplified image generation model that trains fast on ordinary hardware.

NBCE. I made a tool that lets any AI model read documents ten times longer than it was built for.

SimCSE Chinese. I tested a sentence-matching technique across five Chinese datasets.

VAE. I made a simple generative AI model trainer for images.

Word Discovery. I built a tool that finds new Chinese words faster and more accurately than published research.

T5 in Keras. I organized how to use T5 language models in Keras with working examples.

GPLinker. I built a tool that extracts names, relationships, and events from text.

Linear RNN. I implemented three parallel-friendly neural network variants based on recent research.

ON-LSTM. I implemented a neural network that learns sentence structure automatically.

Perturbed Masking. I built an unsupervised tool that breaks Chinese text into words and analyzes sentence structure.

Margin Softmax. I built a tool for training AI models that measure text similarity.

Lookahead. I made a training technique that improves how AI models learn.

Textweave. I built a tool that generates text from unlabeled data.

Flow. I built Keras implementations of three generative flow model architectures.

BERT in Keras. I made simple examples of using BERT to understand text in Keras.

LaBSE. I converted Google's multilingual sentence model into an easy-to-use format.

Entity Linker. I made a model that finds and links entities in text to a database.

Chess Master. I built an AI that plays Chinese chess by learning from real games.

CDial GPT. I built a tool that loads a Chinese conversation AI model in minutes.

Math Problem Solver. I built an AI model that solves elementary math word problems.

Entity Linker. I built a model that connects mentions in text to real entities in a database.

KgCLUE. I built a question answering system that searches knowledge databases.

TextReader. I built a tool that reads Chinese text straight from images.

R-Drop. I built an implementation of R-Drop for training Chinese language AI models.

Infomap. I made a tool that finds natural groups in connected data automatically.

Lazy Optimizer. I made a tool that speeds up neural network training by skipping unused words.

Gradient Accumulator. I built a tool that lets you train AI models with bigger effective batches on limited hardware.

KG-2019. I built an AI model that extracts relationships and facts from text.

Memm. I built a tool that learns patterns in sequences to make predictions.

Beiguo. I made a tool that finds the container running any process by its ID.

CRF. I wrote the shortest implementation of a text sequence labeler.

Pattern Exploiting Training. I built an experiment showing how to train AI models on Chinese text more efficiently.

P-tuning. I experimented with P-tuning to improve how AI models handle Chinese.

RAdam. I built a better training optimizer for Keras AI models.

LIC2020 Baselines. I made reference examples for the LIC2020 language competition.

Word2Vec. I built a faster way to teach computers what words mean.

Shuffle. I made a tool that shuffles hundreds of gigabytes of files with limited memory.