Lakshmi Subramanian

Manuel Tonneau, Neil K. R. Seghal, Niyati Malhotra et al.

Demographic probing is widely used to study how large language models (LLMs) adapt their behavior to signaled demographic attributes. This approach typically uses a single demographic cue in isolation (e.g., a name or dialect) as a signal for group membership, implicitly assuming strong construct validity: that such cues are interchangeable operationalizations of the same underlying, demographically conditioned behavior. We test this assumption in realistic advice-seeking interactions, focusing on race and gender in a U.S. context. We find that cues intended to represent the same demographic group induce only partially overlapping changes in model behavior, while differentiation between groups within a given cue is weak and uneven. Consequently, estimated disparities are unstable, with both magnitude and direction varying across cues. We further show that these inconsistencies partly arise from variation in how strongly cues encode demographic attributes and from linguistic confounders that independently shape model behavior. Together, our findings suggest that demographic probing lacks construct validity: it does not yield a single, stable characterization of how LLMs condition on demographic information, which may reflect a misspecified or fragmented construct. We conclude by recommending the use of multiple, ecologically valid cues and explicit control of confounders to support more defensible claims about demographic effects in LLMs.

Jiali Ma, Jiequan Cui, Maeno Kazuki et al.

In this paper, we propose an Aligned Contrastive Learning (ACL) algorithm to address the long-tailed recognition problem. Our findings indicate that while multi-view training boosts the performance, contrastive learning does not consistently enhance model generalization as the number of views increases. Through theoretical gradient analysis of supervised contrastive learning (SCL), we identify gradient conflicts, and imbalanced attraction and repulsion gradients between positive and negative pairs as the underlying issues. Our ACL algorithm is designed to eliminate these problems and demonstrates strong performance across multiple benchmarks. We validate the effectiveness of ACL through experiments on long-tailed CIFAR, ImageNet, Places, and iNaturalist datasets. Results show that ACL achieves new state-of-the-art performance.

Moumena Chaqfeh, Jacinta Hu, Waleed Hashmi et al.

The amount of JavaScript embedded in Web pages has substantially grown in the past decade, leading to large and complex pages that are computationally intensive for mobile devices. In this paper, we propose JSAnalyzer, an easy-to-use tool that enables Web developers to quickly optimize and generate simpler versions of existing web pages for mobile users. JSAnalyzer can selectively enable or disable JavaScript elements in a page while visually observing their impact, such that non-critical elements can be removed without sacrificing the visual content or the interactive functionality. Our quantitative evaluation results show that JSAnalyzer achieves more than 88% relative increase in performance scoring for low-end mobile phones (i.e., from 32% to 60%), and reduces the page load time by 30%. A qualitative study of 22 users shows that JSAnalyzer maintains more than 90% visual similarity to the original pages, whereas a developer evaluation study conducted with 23 developers shows that JSAnalyzer scores more than 80% in terms of usefulness and usability while retaining the page content and functional features. Additionally, we show that JSAnalyzer outperforms state-of-the-art solutions such as JSCleaner and Google AMP.

Tofunmi Kupoluyi, Moumena Chaqfeh, Matteo Varvello et al.

JavaScript contributes to the increasing complexity of today's web. To support user interactivity and accelerate the development cycle, web developers heavily rely on large general-purpose third-party JavaScript libraries. This practice increases the size and the processing complexity of a web page by bringing additional functions that are not used by the page but unnecessarily downloaded and processed by the browser. In this paper, an analysis of around 40,000 web pages shows that 70% of JavaScript functions on the median page are unused, and the elimination of these functions would contribute to the reduction of the page size by 60%. Motivated by these findings, we propose Muzeel (which means eliminator in Arabic); a solution for eliminating JavaScript functions that are not used in a given web page (commonly referred to as dead code). Muzeel extracts all of the page event listeners upon page load, and emulates user interactions using a bot that triggers each of these events, in order to eliminate the dead code of functions that are not called by any of these events. Our evaluation results spanning several Android mobile phones and browsers show that Muzeel speeds up the page load by around 30% on low-end phones, and by 25% on high-end phones under 3G network. It also reduces the speed index (which is an important user experience metric) by 23% and 21% under the same network on low-end, and high-end phones, respectively. Additionally, Muzeel reduces the overall download size while maintaining the visual content and interactive functionality of the pages.

Ulzee An, Khader Shameer, Lakshmi Subramanian

Applying deep learning methods to mammography assessment has remained a challenging topic. Dense noise with sparse expressions, mega-pixel raw data resolution, lack of diverse examples have all been factors affecting performance. The lack of pixel-level ground truths have especially limited segmentation methods in pushing beyond approximately bounding regions. We propose a classification approach grounded in high performance tissue assessment as an alternative to all-in-one localization and assessment models that is also capable of pinpointing the causal pixels. First, the objective of the mammography assessment task is formalized in the context of local tissue classifiers. Then, the accuracy of a convolutional neural net is evaluated on classifying patches of tissue with suspicious findings at varying scales, where highest obtained AUC is above $0.9$. The local evaluations of one such expert tissue classifier is used to augment the results of a heatmap regression model and additionally recover the exact causal regions at high resolution as a saliency image suitable for clinical settings.