Literature review on Unsupervised Representation Learning in diffusion models.

An interesting paper from ECCV2024. It talks about the color and shape disentanglement on Text-to-Image models. The solution is simple yet effective.

Revisit Contrastive learning Contrastive learning is an instance-wise discriminative approach that aims at making similar instances closer and dissim ...

T-cell receptors (TCR) bind to certain antigens found on abnormal cells. We can perform TCR engineering with VAEs.

ESM-1 and ESM-2 refer to protein language models developed by Facebook AI Research under the project name Evolutionary Scale Modeling (ESM). These models are designed to understand and predict various properties of proteins by leveraging the principles of language modeling in a manner similar to how models like GPT-3 process natural language.

Let's try to figure out whats inside AlphaFold! AlphaFold can accurately predict structures of biomolecular interactions.

ProteinDT is A Opensource Text-guided Protein Design Framework that uses Contrastive learning to align the two modalities.

The Axial Transformer is aimed at managing the complexity of high-dimensional data like images and videos, while The MSA transformer is focused on biological sequences and their alignments.

Identity-Perserving Generation task refers to synthesizing an image that contains a desired person's identity. InstantID tackles this task with only one reference facial image. Moreover, InstantID does not require test-time training while utilizing the learned prior from diffusion-based foundation models like Stable Diffusion. Such methodology is highly efficient compared to previous methods that require fine-tuning on the foundation model. According to the paper, this work shows superior results in Identity-Perserving Generation, outperforming LoRA-based and IP-adapter-based methods qualitatively.

My handwritten notes on Video Diffusion models. Includes TokenFlow, Fairy, LaVie