Performance Analysis
Model Shapley: Equitable Model Valuation with Black-box Access Xinyi Xu, Thanh Lam
ML models call for an equitable model valuation method to price them. In particular, we investigate the black-box access setting which allows querying a model (to observe predictions) without disclosing model-specific information (e.g., architecture and parameters). By exploiting a Dirichlet abstraction of a model's predictions, we propose a novel and equitable model valuation method called
Analysis of Variance of Multiple Causal Networks
Constructing a directed cyclic graph (DCG) is challenged by both algorithmic difficulty and computational burden. Comparing multiple DCGs is even more difficult, compounded by the need to identify dynamic causalities across graphs.
Achievable Fairness on Y our Data With Utility Guarantees
Secondly, these works do not account for finite-sampling errors in the obtained curve.
No Free Lunch in LLM Watermarking: Trade-offs in Watermarking Design Choices
Advances in generative models have made it possible for AI-generated text, code, and images to mirror human-generated content in many applications. W atermark-ing, a technique that aims to embed information in the output of a model to verify its source, is useful for mitigating the misuse of such AI-generated content. However, we show that common design choices in LLM watermarking schemes make the resulting systems surprisingly susceptible to attack--leading to fundamental trade-offs in robustness, utility, and usability. To navigate these trade-offs, we rigorously study a set of simple yet effective attacks on common watermarking systems, and propose guidelines and defenses for LLM watermarking in practice.
Human Expertise in Algorithmic Prediction
Algorithmic indistinguishability yields a natural test for assessing whether experts incorporate this kind of "side
Divergences between Language Models and Human Brains
Do machines and humans process language in similar ways?
Graph Classification via Reference Distribution Learning: Theory and Practice
This work introduces Graph Reference Distribution Learning (GRDL), an efficient and accurate graph classification method. GRDL treats each graph's latent node embeddings given by GNN layers as a
A Disparity Metric Definitions 566 A.1 Observational Metrics
U 2 U that influences all of the variables U influences. Figure 5: Example of step one in the marginalisation, taken from Evans [22]. In this section we analyse the datasets presented in Le Quy et al. For each bias we provide a justification of our decision. Therefore we drop them from the analysis. Diabetes For this dataset, the goal is to predict if a patient will be readmitted in the next 30 days.