Topics of interest

Topics of interest include (but are not limited to):

Robustness and reliability

adversarial attacks against TFMs (e.g. context manipulation, distribution shift), defenses (e.g. adaptation, fine-tuning, outlier detection), missing or corrupted data, calibration, uncertainty, and reliable adaptation or fine-tuning;

Privacy and security

membership inference, data leakage from inference-time contexts, data/model extraction and reconstruction attacks, poisoning, and privacy-preserving inference;

Fairness and responsible prediction

group and individual fairness, context-induced disparities, counterfactual fairness, and fairness under distribution shift;

Interpretability and explainability

feature attribution and post-hoc explanations, robustness and stability of explanations, analysis of representations and attention, invariance properties, auditing, understanding the roles of the pretrained prior and inference-time context, and identifying the influence of specific support examples;

Evaluation and guarantees

trustworthy benchmarks, tools and demos in real-world applications, threat models, reproducibility, certification, verification, and evaluation in high-stakes settings.