Gaurush Hiranandani

🚀 New Research Alert: Logits are All We Need to Adapt Closed Models 🔒Many commercial Large Language Models (LLMs), e.g., GPT-4, are closed-source, limiting developers to steer content generation. 🤔Can we adapt closed-source LLMs when fine-tuning or accessing their internal weights is not possible? Check out our work by @gaurushh, @Haolun_Wu0203, Subhojyoti, @sanmikoyejo from Stanford @stai_research. 1/n

Excited to share our new work on black-box metric optimization, published at ICML'21. @sanmikoyejo Paper Link: proceedings.mlr.press/v139/hirananda… Source Code: github.com/koyejolab/fweg

We are excited to discuss our work at NeurIPS 2020. Read the paper for more details here arxiv.org/abs/2006.12732. We are glad to hear your comments and feedback. (n/n) #NeurIPS2020 #fairness #MachineLearning

Specifically, our approach elicits fair performance metrics for multiclass-multigroup classification problems, including selecting the trade-off between predictive performance and fairness violation using only a few pairwise preference comparisons. (4/n)

choose and customize metrics that reflect their belief of what constitutes a fair classifier. (3/n)

End-users often face the dilemma of identifying the best among the bouquet of metrics to measure and mitigate these biases. In our recent work with Harikrishna Narasimhan and @sanmikoyejo, we have developed a novel metric elicitation method to help end-users (2/n)

Over the years, algorithmic decision making has evolved to be more complex, yet existing algorithms are susceptible to biases, where decisions may be unfair to specific population subgroups. (1/n)