Influence Graph

Influence graphs represent the relationships between data points, revealing how individual data items impact others and the overall system. Current research focuses on learning these graphs from various data types (e.g., social networks, time series) using algorithms that leverage influence functions and graph embeddings to identify influential data points and disentangle causal relationships. This work has significant implications for improving model robustness (e.g., by detecting and mitigating backdoor attacks), enhancing recommendation systems, and understanding the spread of misinformation by analyzing the impact of individual data points on model predictions and outcomes.