arXiv subjects

Sumit Agarwal

Publications and source records attributed to Sumit Agarwal.

7 recordsLinked to original sources

arXiv2025

Cash Transfers in the Perinatal Period and Child Welfare System Involvement Among Infants: Evidence from the Rx Kids Program in Flint, Michigan

Infants are most vulnerable to child maltreatment, which may be due in part to economic instability during the perinatal period. In 2024, Rx Kids was launched in Flint, Michigan, achieving near 100% aggregate take up and providing every expectant mother with unconditional cash transfers during pregnancy and infancy. Synthetic difference-in-differences was used to compare changes in allegations of maltreatment within the first six months of life in Flint before and after implementation of Rx Kids relative to the corresponding change in control cities without the program. In the three years prior to the implementation of Rx Kids, the proportion of infants with a maltreatment allegation within the first six months of life was 21.7% in Flint and 19.5% among control cities. After implementation of Rx Kids in 2024, the maltreatment allegation rate dropped to 15.5% in Flint, falling below the maltreatment allegation rate of 20.6% among the control cities. Rx Kids was associated with a statistically significant 7.0 percentage-point decrease in the maltreatment allegation rate (p = 0.021), corresponding to a 32% decrease relative to the pre-intervention period. There was a decrease in the rate of neglect-related, non-neglect-related, and substantiated allegations; these were directionally consistent with the primary outcome but not statistically significant. Results were robust to alternative model specifications. The Rx Kids prenatal and infant cash prescription program led to a significant reduction in allegations of maltreatment among infants. These findings provide important evidence about the role of economic stability in preventing child welfare system involvement.

econ.GN↗

arXiv2024

Transportation Technology and Gentrification: Evidence from the entry of Ridesharing Services

We analyze the staggered entry of rideshare services across U.S. metropolitan areas, estimating its effect on the spatial redistribution and real outcomes of residents. Ridesharing services gentrify urban areas-especially those with ex-ante lower housing values-causing housing prices to rise 9 percent, with the in-migration of rich-younger individuals more than offsetting the out-migration of incumbent residents and reduced in-migration of poorer individuals. Impact on incumbent residents is conditional on ex-ante homeownership. For homeowners, there is no displacement and a decline in delinquency rates. For non-homeowners, displacement and delinquency rates rise 11 percent and 42 percent, respectively. Our study emphasizes how the private provision of high-end transportation technologies can increase urbanization and exacerbate inequality.

econ.GN↗

arXiv2023

PEFTDebias : Capturing debiasing information using PEFTs

The increasing use of foundation models highlights the urgent need to address and eliminate implicit biases present in them that arise during pretraining. In this paper, we introduce PEFTDebias, a novel approach that employs parameter-efficient fine-tuning (PEFT) to mitigate the biases within foundation models. PEFTDebias consists of two main phases: an upstream phase for acquiring debiasing parameters along a specific bias axis, and a downstream phase where these parameters are incorporated into the model and frozen during the fine-tuning process. By evaluating on four datasets across two bias axes namely gender and race, we find that downstream biases can be effectively reduced with PEFTs. In addition, we show that these parameters possess axis-specific debiasing characteristics, enabling their effective transferability in mitigating biases in various downstream tasks. To ensure reproducibility, we release the code to do our experiments.

cs.LG↗

arXiv2023

CodeBERTScore: Evaluating Code Generation with Pretrained Models of Code

Since the rise of neural natural-language-to-code models (NL->Code) that can generate long expressions and statements rather than a single next-token, one of the major problems has been reliably evaluating their generated output. In this paper, we propose CodeBERTScore: an evaluation metric for code generation, which builds on BERTScore (Zhang et al., 2020). Instead of encoding only the generated tokens as in BERTScore, CodeBERTScore also encodes the natural language input preceding the generated code, thus modeling the consistency between the generated code and its given natural language context as well. We perform an extensive evaluation of CodeBERTScore across four programming languages. We find that CodeBERTScore achieves a higher correlation with human preference and with functional correctness than all existing metrics. That is, generated code that receives a higher score by CodeBERTScore is more likely to be preferred by humans, as well as to function correctly when executed. We release five language-specific pretrained models to use with our publicly available code. Our language-specific models have been downloaded more than 1,000,000 times from the Huggingface Hub. Our code and data are available at https://github.com/neulab/code-bert-score

cs.SE↗

arXiv2019

Illumination-invariant Face recognition by fusing thermal and visual images via gradient transfer

Face recognition in real life situations like low illumination condition is still an open challenge in biometric security. It is well established that the state-of-the-art methods in face recognition provide low accuracy in the case of poor illumination. In this work, we propose an algorithm for a more robust illumination invariant face recognition using a multi-modal approach. We propose a new dataset consisting of aligned faces of thermal and visual images of a hundred subjects. We then apply face detection on thermal images using the biggest blob extraction method and apply them for fusing images of different modalities for the purpose of face recognition. An algorithm is proposed to implement fusion of thermal and visual images. We reason for why relying on only one modality can give erroneous results. We use a lighter and faster CNN model called MobileNet for the purpose of face recognition with faster inferencing and to be able to be use it in real time biometric systems. We test our proposed method on our own created dataset to show that real-time face recognition on fused images shows far better results than using visual or thermal images separately.

cs.CV↗

arXiv2017

Location Optimization of ATM Networks

ATMs enable the public to perform financial transactions. Banks try to strategically position their ATMs in order to maximize transactions and revenue. In this paper, we introduce a model which provides a score to an ATM location, which serves as an indicator of its relative likelihood of transactions. In order to efficiently capture the spatially dynamic features, we utilize two concurrent prediction models: the local model which encodes the spatial variance by considering highly energetic features in a given location, and the global model which enforces the dominant trends in the entire data and serves as a feedback to the local model to prevent overfitting. The major challenge in learning the model parameters is the lack of an objective function. The model is trained using a synthetic objective function using the dominant features returned from the k-means clustering algorithm in the local model. The results obtained from the energetic features using the models are encouraging.

cs.CY↗

arXiv2014

Design and Analysis of RS Sort

This paper introduces a new comparison base stable sorting algorithm, named RS sort. RS Sort involves only the comparison of pair of elements in an array which ultimately sorts the array and does not involve the comparison of each element with every other element. RS sort tries to build upon the relationship established between the elements in each pass. Suppose there is an array containing three elements a1, a2, a3 and if a relationship exist such that a1<a2 and a2<a3 then it can be established that a1<a3 and so there is no need to compare a1 and a3. Sorting is a fundamental operation in computer science. RS sort is analyzed both theoretically and empirically. We have performed its Empirical analysis and compared its performance with the well-known quick sort for various input types.

cs.DS↗