Search AI Products and News

Artificial intelligence is rapidly transforming protein discovery and design in biotechnology. Recently, a collaborative team from UC Berkeley and Caltech developed ProteinDT, a novel multimodal framework leveraging text descriptions to aid in protein design. This innovative approach integrates protein sequence and structural information with a wealth of biological knowledge represented in text, opening a new chapter in protein design.

Researchers at the University of California, Berkeley have recently released their latest research achievement – the TULIP (Towards Unified Language-Image Pretraining) model. This model aims to enhance the performance of vision-language pre-training, particularly in visually-centric tasks requiring high-fidelity understanding, overcoming the limitations of existing contrastive learning models (such as CLIP). TULIP integrates innovative techniques such as generative data augmentation, enhanced contrastive learning, and reconstruction regularization.

Recently, Tencent Technology (Shenzhen) Co., Ltd. published a patent regarding a training method and related equipment for large language models on the Tianyancha app. The patent is titled 'Training Method, Device, Computer Equipment, and Storage Medium for Large Language Models' and aims to enhance the learning capacity and accuracy of large language models through innovative training methods. In the training process of large language models, traditional methods often rely on a single text summary, which may lead to model overfitting and negatively impact the accuracy and diversity of generated content. However, Tencent's new...

In today's technology landscape, CLIP (Contrastive Language-Image Pre-training) is an important multimodal foundational model. It combines visual signals and text signals into a shared feature space using contrastive learning loss on a large-scale dataset of image-text pairs. As a retriever, CLIP supports various tasks such as zero-shot classification, detection, segmentation, and image-text retrieval. Meanwhile, as a feature extractor, it performs well in nearly all...

EasyRec is a recommendation system based on language models, developed by a team from the University of Hong Kong. Its uniqueness lies in analyzing emotional and detailed user behavior stories through a text behavior alignment framework to predict user preferences without requiring large amounts of user data. The system combines contrastive learning and collaborative language models, enabling accurate predictions of preferences for new users and new products, particularly excelling in zero-shot recommendation scenarios. EasyRec's plug-and-play features make it easy to integrate into existing recommendation systems, enhancing performance. The paper showcases EasyRec's performance across multiple...

Researchers from MIT and Google have collaborated to develop StableRep technology, which trains detailed and efficient AI image models using AI-generated images. StableRep employs multi-positive contrastive learning methods and trains on millions of labeled synthetic images, achieving significant results in ImageNet classification. Despite its success, StableRep's image generation is relatively slow, suffers from semantic mismatch issues, and requires the underlying model to be initially trained on real data. The technology has been open-sourced on GitHub.

The latest research from Renmin University warns that caution is needed in data augmentation for contrastive learning. Strongly aligned positive samples may harm the generalization ability of contrastive learning. While stronger data augmentation can enhance downstream task performance, alignment performance decreases. The study reveals the mechanisms by which data augmentation affects contrastive learning and proposes seeking better data augmentation strategies from an information-theoretic and spectral perspective.