Search

Increasing use of large language models (LLMs) demand performant guardrails to ensure the safety of inputs and outputs of LLMs. When these safeguards are trained on imbalanced data, they can learn the societal biases. We present a light-weight, post-processing method for mitigating counterfactual fairness in closed-source text safety classifiers. Our approach involves building an ensemble that not only outperforms the input classifiers and policy-aligns them, but also acts as a debiasing regularizer. We introduce two threshold-agnostic metrics to assess the counterfactual fairness of a model, and demonstrate how combining these metrics with Fair Data Reweighting (FDW) helps mitigate biases. We create an expanded Open AI dataset, and a new templated LLM-generated dataset based on user-prompts, both of which are counterfactually balanced across identity groups and cover four key areas of safety; we will work towards publicly releasing these datasets. Our results show that our approach improves counterfactual fairness with minimal impact on model performance.

Given the growing prevalence of diabetes, there has been significant interest in determining how diabetes affects instrumental daily functions, like driving. Complication of glucose control in diabetes includes hypoglycemic and hyperglycemic episodes, which may impair cognitive and psychomotor functions needed for safe driving. The goal of this paper was to determine patterns of diabetes speed behavior during acute glucose to drivers with diabetes who were euglycemic or control drivers without diabetes in a naturalistic driving environment. By employing distribution-based analytic methods which capture distribution patterns, our study advances prior literature that has focused on conventional approach of average speed to explore speed deviation patterns., Comment: 6 pages, 2 figures, 5 Tables, Accepted and Presented at IEEE ITSC 2023 Conference in Bilbao Spain

As the use of deep learning in high impact domains becomes ubiquitous, it is increasingly important to assess the resilience of models. One such high impact domain is that of face recognition, with real world applications involving images affected by various degradations, such as motion blur or high exposure. Moreover, images captured across different attributes, such as gender and race, can also challenge the robustness of a face recognition algorithm. While traditional summary statistics suggest that the aggregate performance of face recognition models has continued to improve, these metrics do not directly measure the robustness or fairness of the models. Visual Psychophysics Sensitivity Analysis (VPSA) [1] provides a way to pinpoint the individual causes of failure by way of introducing incremental perturbations in the data. However, perturbations may affect subgroups differently. In this paper, we propose a new fairness evaluation based on robustness in the form of a generic framework that extends VPSA. With this framework, we can analyze the ability of a model to perform fairly for different subgroups of a population affected by perturbations, and pinpoint the exact failure modes for a subgroup by measuring targeted robustness. With the increasing focus on the fairness of models, we use face recognition as an example application of our framework and propose to compactly visualize the fairness analysis of a model via AUC matrices. We analyze the performance of common face recognition models and empirically show that certain subgroups are at a disadvantage when images are perturbed, thereby uncovering trends that were not visible using the model's performance on subgroups without perturbations., Comment: 8 pages, 5 figures, to be published in NeurIPS 2021 Workshop, Algorithmic Fairness through the Lens of Causality and Robustness

The ability to automatically detect stuttering events in speech could help speech pathologists track an individual's fluency over time or help improve speech recognition systems for people with atypical speech patterns. Despite increasing interest in this area, existing public datasets are too small to build generalizable dysfluency detection systems and lack sufficient annotations. In this work, we introduce Stuttering Events in Podcasts (SEP-28k), a dataset containing over 28k clips labeled with five event types including blocks, prolongations, sound repetitions, word repetitions, and interjections. Audio comes from public podcasts largely consisting of people who stutter interviewing other people who stutter. We benchmark a set of acoustic models on SEP-28k and the public FluencyBank dataset and highlight how simply increasing the amount of training data improves relative detection performance by 28\% and 24\% F1 on each. Annotations from over 32k clips across both datasets will be publicly released., Comment: Accepted to ICASSP 2021

We consider a novel setting of zeroth order non-convex optimization, where in addition to querying the function value at a given point, we can also duel two points and get the point with the larger function value. We refer to this setting as optimization with dueling-choice bandits since both direct queries and duels are available for optimization. We give the COMP-GP-UCB algorithm based on GP-UCB (Srinivas et al., 2009), where instead of directly querying the point with the maximum Upper Confidence Bound (UCB), we perform a constrained optimization and use comparisons to filter out suboptimal points. COMP-GP-UCB comes with theoretical guarantee of $O(\frac{\Phi}{\sqrt{T}})$ on simple regret where $T$ is the number of direct queries and $\Phi$ is an improved information gain corresponding to a comparison based constraint set that restricts the search space for the optimum. In contrast, in the direct query only setting, $\Phi$ depends on the entire domain. Finally, we present experimental results to show the efficacy of our algorithm., Comment: 19 pages, 3 figures

Subgraph discovery in a single data graph---finding subsets of vertices and edges satisfying a user-specified criteria---is an essential and general graph analytics operation with a wide spectrum of applications. Depending on the criteria, subgraphs of interest may correspond to cliques of friends in social networks, interconnected entities in RDF data, or frequent patterns in protein interaction networks to name a few. Existing systems usually examine a large number of subgraphs while employing many computers and often produce an enormous result set of subgraphs. How can we enable fast discovery of only the most relevant subgraphs while minimizing the computational requirements? We present Nuri, a general subgraph discovery system that allows users to succinctly specify subgraphs of interest and criteria for ranking them. Given such specifications, Nuri efficiently finds the k most relevant subgraphs using only a single computer. It prioritizes (i.e., expands earlier than others) subgraphs that are more likely to expand into the desired subgraphs (prioritized subgraph expansion) and proactively discards irrelevant subgraphs from which the desired subgraphs cannot be constructed (pruning). Nuri can also efficiently store and retrieve a large number of subgraphs on disk without being limited by the size of main memory. We demonstrate using both real and synthetic datasets that Nuri on a single core outperforms the closest alternative distributed system consuming 40 times more computational resources by more than 2 orders of magnitude for clique discovery and 1 order of magnitude for subgraph isomorphism and pattern mining.