Advances in Biomedical AI and Watermarking

The field of biomedical AI is moving towards the development of foundation models that can integrate multimodal biological data and perform a wide range of machine learning tasks. These models have the potential to revolutionize the field of drug discovery and translational research. Recent studies have shown that multimodal deep learning frameworks can be used to detect biomedical misconduct and predict drug response in single-cell data. Additionally, the development of open-source machine learning platforms has enabled the streamlined training, evaluation, and inference of models that integrate multimodal biological data.

In the area of watermarking, researchers are exploring new methods to embed watermarks in AI-generated text and images. These methods include the use of Lagrange interpolation, semantic information, and optimized couplings. The development of robust watermarking techniques is essential for combating misinformation and protecting intellectual property rights.

Some noteworthy papers in this area include PyTDC, which provides a platform for training, evaluating, and inferring multimodal biological AI models, and scDrugMap, which benchmarks large foundation models for drug response prediction in single-cell data. The paper on watermarking using Lagrange interpolation presents a highly effective method for recovering author identity, while the study on removing watermarks with partial regeneration using semantic information highlights the need for more robust watermarking algorithms.

Sources

PyTDC: A multimodal machine learning training, evaluation, and inference platform for biomedical foundation models

scDrugMap: Benchmarking Large Foundation Models for Drug Response Prediction

LLM-Text Watermarking based on Lagrange Interpolation

BMMDetect: A Multimodal Deep Learning Framework for Comprehensive Biomedical Misconduct Detection

RAP-SM: Robust Adversarial Prompt via Shadow Models for Copyright Verification of Large Language Models

Sandcastles in the Storm: Revisiting the (Im)possibility of Strong Watermarking

CNN-based Image Models Verify a Hypothesis that The Writers of Cuneiform Texts Improved Their Writing Skills When Studying at the Age of Hittite Empire

Characterizing the Investigative Methods of Fictional Detectives with Large Language Models

Multimodal Survival Modeling in the Age of Foundation Models

Removing Watermarks with Partial Regeneration using Semantic Information

Human-AI Collaboration or Academic Misconduct? Measuring AI Use in Student Writing Through Stylometric Evidence

Optimized Couplings for Watermarking Large Language Models

From Trade-off to Synergy: A Versatile Symbiotic Watermarking Framework for Large Language Models

Built with on top of