Overview This paper introduces a comprehensive review of knowledge graph. There are four topics about KG, 1) knowledge graph representation learning, 2) knowledge acquisition and completion, 3) tem...

Conclusion By Myself This is a really clearly paper. It introduces the LLaMA. LLaMA’s training approach is similar to the methods described in Brown et al., 2020;Chowdhery et al., 2020, and is ins...

2023-11-07-Deep Residual Learning for Image Recognition

Abstraction Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. W...

Conclusion By Myself The traditional technology of word representation is use an extra model to train a stable word vectors. ELMo can give a dynamic word representations by a language model. This L...

Conclusion By Myself This paper introduced a Multimodal LLM, which combine image tasks and text tasks together. It use CLIP ViT-L/14 model to representing the pic, and use a transformer based model...

Conclusion By Myself Feed more related data to model at once can significantly enhance its performance across multiple aspects. Use the cosine similarity to determine the similarity between any two...

Conclusion By Myself LLMs lack of the ability to write modularized code to solve complex task. This paper introduced a way of inference to solve this problem. Concretely in two step: Use “chain-...

Conclusion By Myself This is a paper from ByteDance, it presented an open-source and reproducible evaluation suite for LLMs named GPT-Fathom. Abstract With the rapid advancement of LLMs, there is a...

Conclusion By Myself This paper introduces CodeFusion, which is a pre-trained diffusion model for code generation. It combines an encoder-decoder architecture with a diffusion process. In more deta...