Search

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.

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., Comment: EMNLP 2023

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

To address potential gaps noted in patient monitoring in the hospital, a novel patient behavior detection system using mmWave radar and deep convolution neural network (CNN), which supports the simultaneous recognition of multiple patients' behaviors in real-time, is proposed. In this study, we use an mmWave radar to track multiple patients and detect the scattering point cloud of each one. For each patient, the Doppler pattern of the point cloud over a time period is collected as the behavior signature. A three-layer CNN model is created to classify the behavior for each patient. The tracking and point clouds detection algorithm was also implemented on an mmWave radar hardware platform with an embedded graphics processing unit (GPU) board to collect Doppler pattern and run the CNN model. A training dataset of six types of behavior were collected, over a long duration, to train the model using Adam optimizer with an objective to minimize cross-entropy loss function. Lastly, the system was tested for real-time operation and obtained a very good inference accuracy when predicting each patient's behavior in a two-patient scenario., Comment: This paper has been submitted to IEEE Radar Conference 2019

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.

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., Comment: 5 pages; 2 figures; 2 tables

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

We study credit card rewards as an ideal laboratory to quantify redistribution between consumers in retail financial markets. Comparing cards with and without rewards, we find that, regardless of income, sophisticated individuals profit from reward credit cards at the expense of na¨ıve consumers. To probe the underlying mechanisms, we exploit bank-initiated account limit increases at the card level and show that reward cards induce more spending, leaving na¨ıve consumers with higher unpaid balances. Na¨ıve consumers also follow a sub-optimal balance-matching heuristic when repaying their credit cards, incurring higher costs. Banks incentivize the use of reward cards by offering lower interest rates than on comparable cards without rewards. We estimate an aggregate annual redistribution of $15 billion from less to more educated, poorer to richer, and high to low minority areas, widening existing disparities. [ABSTRACT FROM AUTHOR]

A large-scale microcredit expansion program--together with a credit bureau accessible to all lenders--can enable unbanked borrowers to build a credit history, facilitating their transition to commercial banks. Loan-level data from Rwanda show the program improved access to credit and reduced poverty. A sizable share of first-time borrowers switched to commercial banks, which cream-skim less risky borrowers and grant them larger, cheaper, and longer-maturity loans. Switchers have lower default risk than non-switchers and are not riskier than other bank borrowers. Switchers also obtain better loan terms from banks compared with first-time bank borrowers without a credit history. [ABSTRACT FROM AUTHOR]

How does a bank react when a substantial share of its borrowers suffer a large negative shock? To answer this question we exploit the 2014 collapse of energy prices using the universe of Mexican commercial bank loans. We show that, after the drop in energy prices, banks exposed to the energy sector increased their exposure to these borrowers even more, relaxing credit margins to their larger debtors in the sector. An increase of one standard deviation in a bank's ex-ante exposure to the energy sector increased the loan volume to borrowers in the sector by 18 percent and reduced interest rates by 6 percent, even though borrower's credit default swap spreads were widening. Highly exposed banks amplified this sector-specific shock to the rest of the economy by contracting lending to other sectors, with important real effects, as the borrowers could not switch credit suppliers. Finally, the energy price shock had a large negative impact on macro outcomes, especially in the capital-intensive secondary sector. Quantitatively, a one standard deviation increase in the exposure of a state's banks to the energy sector reduced its GDP by 1.8 percent. [ABSTRACT FROM AUTHOR]