A5027454515- Privacy-Preserving Technologies in Data
- Topic Modeling
- Poverty, Education, and Child Welfare
- Stochastic Gradient Optimization Techniques
- Artificial Intelligence in Healthcare and Education
- School Choice and Performance
- Machine Learning in Healthcare
- Gender Diversity and Inequality
- Income, Poverty, and Inequality
- COVID-19 Pandemic Impacts
- Natural Language Processing Techniques
- Domain Adaptation and Few-Shot Learning
- Experimental Behavioral Economics Studies
- Recommender Systems and Techniques
- Global Educational Reforms and Inequalities
- Microfinance and Financial Inclusion
- Multimodal Machine Learning Applications
- Cancer-related molecular mechanisms research
- Labor market dynamics and wage inequality
- Biomedical Text Mining and Ontologies
- Social and Economic Development in India
- Child Nutrition and Water Access
- Innovations in Educational Methods
- Migration and Labor Dynamics
- Gender Politics and Representation
Luxembourg Institute of Socio-Economic Research
2017-2025
University of Luxembourg
2017-2025
Google (United States)
2019-2024
University of Arizona
2023
Indian Institute of Management Ahmedabad
2018-2022
Stanford University
2019
Abstract Large language models (LLMs) have demonstrated impressive capabilities, but the bar for clinical applications is high. Attempts to assess knowledge of typically rely on automated evaluations based limited benchmarks. Here, address these limitations, we present MultiMedQA, a benchmark combining six existing medical question answering datasets spanning professional medicine, research and consumer queries new dataset questions searched online, HealthSearchQA. We propose human...
Recent artificial intelligence (AI) systems have reached milestones in "grand challenges" ranging from Go to protein-folding. The capability retrieve medical knowledge, reason over it, and answer questions comparably physicians has long been viewed as one such grand challenge. Large language models (LLMs) catalyzed significant progress question answering; Med-PaLM was the first model exceed a "passing" score US Medical Licensing Examination (USMLE) style with of 67.2% on MedQA dataset....
Federated learning and analytics are a distributed approach for collaboratively models (or statistics) from decentralized data, motivated by designed privacy protection. The process can be formulated as solving federated optimization problems, which emphasize communication efficiency, data heterogeneity, compatibility with system requirements, other constraints that not primary considerations in problem settings. This paper provides recommendations guidelines on formulating, designing,...
At the heart of medicine lies physician-patient dialogue, where skillful history-taking paves way for accurate diagnosis, effective management, and enduring trust. Artificial Intelligence (AI) systems capable diagnostic dialogue could increase accessibility, consistency, quality care. However, approximating clinicians' expertise is an outstanding grand challenge. Here, we introduce AMIE (Articulate Medical Explorer), a Large Language Model (LLM) based AI system optimized dialogue. uses novel...
An accurate differential diagnosis (DDx) is a cornerstone of medical care, often reached through an iterative process interpretation that combines clinical history, physical examination, investigations and procedures. Interactive interfaces powered by Large Language Models (LLMs) present new opportunities to both assist automate aspects this process. In study, we introduce LLM optimized for diagnostic reasoning, evaluate its ability generate DDx alone or as aid clinicians. 20 clinicians...
Personalization methods in federated learning aim to balance the benefits of and local training for data availability, communication cost, robustness client heterogeneity. Approaches that require clients communicate all model parameters can be undesirable due privacy constraints. Other approaches always-available or stateful clients, impractical large-scale cross-device settings. We introduce Federated Reconstruction, first model-agnostic framework partially suitable inference at scale....
Large language models (LLMs) have demonstrated impressive capabilities in natural understanding and generation, but the quality bar for medical clinical applications is high. Today, attempts to assess models' knowledge typically rely on automated evaluations limited benchmarks. There no standard evaluate model predictions reasoning across a breadth of tasks. To address this, we present MultiMedQA, benchmark combining six existing open question answering datasets spanning professional exams,...
A recent trend in artificial intelligence is the use of pretrained models for language and vision tasks, which have achieved extraordinary performance but also puzzling failures. Probing these models' abilities diverse ways therefore critical to field. In this paper, we explore reliability models, where define a reliable model as one that not only achieves strong predictive performs well consistently over many decision-making tasks involving uncertainty (e.g., selective prediction, open set...
Medicine is inherently multimodal, with rich data modalities spanning text, imaging, genomics, and more. Generalist biomedical artificial intelligence (AI) systems that flexibly encode, integrate, interpret this at scale can potentially enable impactful applications ranging from scientific discovery to care delivery. To the development of these models, we first curate MultiMedBench, a new multimodal benchmark. MultiMedBench encompasses 14 diverse tasks such as medical question answering,...
This chapter presents evidence of the challenges faced by women and underrepresented minorities in Economics. It, first, examines demographics economics profession, highlighting significant disparities representation. Despite some progress, under representation remains prevalent at different educational levels higher academic positions, for most part. Subsequently, reviews research on existing barriers biases contributing to this Recent work has emphasized crucial role attitudes...
Does it pay to be beautiful?Physically attractive people can earn more, particularly in customerfacing jobs, and the rewards for men are higher than women
Existing literature has established that a diverse workforce is more creative and productive, with academia being no exception. Research on gender diversity in academia, especially economics so far focused the developed world. This article examines India by analyzing share of women faculty positions, journal publications, participation conference held annually since 2004. Unlike some countries, students actually constitute majority at Master's level India. Yet, evidence suggests women's...
We study the relative effects of data augmentations, pretraining algorithms, and model architectures in Self-Supervised Learning (SSL). While recent literature this space leaves impression that algorithm is critical importance to performance, understanding its effect complicated by difficulty making objective direct comparisons between methods. propose a new framework which unifies many seemingly disparate SSL methods into single shared template. Using framework, we identify aspects differ...
Conventional federated learning algorithms train a single global model by leveraging all participating clients’ data. However, due to heterogeneity in client generative distributions and predictive models, these approaches may not appropriately approximate the process, converge an optimal state, or generalize new clients. We study personalization generalization stateless cross-device setups assuming data models. first propose hierarchical formalize it using Bayesian Inference. then this...
Federated learning data is drawn from a distribution of distributions: clients are meta-distribution, and their local distributions. Thus generalization studies in federated should separate performance gaps unseen client (out-of-sample gap) distributions (participation gap). In this work, we propose framework for disentangling these gaps. Using framework, observe explain differences behavior across natural synthetic datasets, indicating that dataset synthesis strategy can be important...
In the view of increasing preference private schooling in India, this article assesses its impact on learning outcomes for rural children from 8 to 11 years. Despite earlier attempts study issue, stands out two ways. Firstly, it addresses problems arising because non-random selection attending schools. Secondly, also presents an all-India estimate unlike most studies which have dealt largely with few states. Our results show performance school be significantly better than those public This...
Using multiple datasets on learning outcomes of over 2 million children in the age group 8–16 years from rural India, paper examines prevalence and persistence gender gap performance mathematics. We observe significant persistent mathematics, which is not found reading outcomes. The northern states lag behind southern ones show a ‘reverse gap’. Pre-existing household practices societal norms are to explain this difference significantly. findings call for need understand these gaps granularly...
Federated learning (FL) enables from decentralized privacy-sensitive data, with computations on raw data confined to take place at edge clients. This paper introduces mixed FL, which incorporates an additional loss term calculated the coordinating server (while maintaining FL's private restrictions). There are numerous benefits. For example, datacenter can be leveraged jointly learn centralized (datacenter) and (federated) training better match expected inference distribution. Mixed FL also...
There has been significant interest recently in learning multilingual word embeddings -- which semantically similar words across languages have embeddings. State-of-the-art approaches relied on expensive labeled data, is unavailable for low-resource languages, or involved post-hoc unification of monolingual In the present paper, we investigate efficacy learned from weakly-supervised image-text data. particular, propose methods using by enforcing similarity between representations image and...
Large language models (LLMs) hold immense promise to serve complex health information needs but also have the potential introduce harm and exacerbate disparities. Reliably evaluating equity-related model failures is a critical step toward developing systems that promote equity. In this work, we present resources methodologies for surfacing biases with precipitate harms in long-form, LLM-generated answers medical questions then conduct an empirical case study Med-PaLM 2, resulting largest...
Can inequalities in private school access be bridged through a government mandate? Enacted 2009, India's "Right to Education" mandated almost all schools admit at least 25 percent of children their entry class from "economically weak and socially disadvantaged" groups. In this paper, we investigate the impact mandate on nature chosen by targeted households one largest cities India. Applying double-difference estimation strategy, compare choices elder siblings (not eligible for mandate)...
We analyze findings from a large-scale survey of over 11,000 respondents across 64 districts in India, conducted between December 2020 and January 2021 to examine the impact lockdown on internal migrants India. find that compared households without migrants, with were relatively advantaged income levels before pandemic but faced more severe food financial vulnerability even nine months after first lockdown. In addition, governmental social security support was difficult access for migrants....