A5100658205- Topic Modeling
- Computational Geometry and Mesh Generation
- Advanced Combinatorial Mathematics
- Software Engineering Research
- Advanced Graph Theory Research
- Cloud Computing and Resource Management
- Machine Learning and Data Classification
- Explainable Artificial Intelligence (XAI)
- Complexity and Algorithms in Graphs
- Blockchain Technology Applications and Security
- Service-Oriented Architecture and Web Services
- Graph theory and applications
- Cryptography and Data Security
- Privacy, Security, and Data Protection
- Software Reliability and Analysis Research
- Advanced Data Storage Technologies
- Game Theory and Applications
- Random Matrices and Applications
- State Capitalism and Financial Governance
- Cryptographic Implementations and Security
- Parallel Computing and Optimization Techniques
- Corporate Finance and Governance
- DNA and Biological Computing
- Privacy-Preserving Technologies in Data
- Bayesian Methods and Mixture Models
University of Washington
2022-2024
California University of Pennsylvania
2024
University of California, San Diego
2023
Fordham University
2023
Massachusetts Institute of Technology
2018-2022
Western University
2022
Stanford University
2021
Genmab (United States)
2020
Iowa State University
2020
University of California, Berkeley
2019
We demonstrate that a neural network pretrained on text and fine-tuned code solves mathematics course problems, explains solutions, generates questions at human level. automatically synthesize programs using few-shot learning OpenAI's Codex transformer execute them to solve problems 81% automatic accuracy. curate dataset of from Massachusetts Institute Technology (MIT)'s largest courses (Single Variable Multivariable Calculus, Differential Equations, Introduction Probability Statistics,...
Automated program repair (APR) aims to help developers improve software reliability by generating patches for buggy programs. Although many code language models (CLM) are developed and effective in tasks such as completion, there has been little comprehensive, in-depth work evaluate CLMs' fixing capabilities fine-tune CLMs the APR task. Firstly, this is first ten on four benchmarks, which shows that surprisingly, best CLM, is, fixes 72% more bugs than state-of-the-art deep-learning...
Software systems are now ubiquitous and used every day for automation purposes in personal enterprise applications; they also essential to many safety-critical mission-critical systems, e.g., air traffic control autonomous cars, SCADA systems. With the availability of massive storage capabilities, high speed Internet, advent Internet Things devices, modern software growing both size complexity. Maintaining a quality such complex while manually keeping error rate at minimum is challenge....
We explore the performance of latent variable models for conditional text generation in context neural machine translation (NMT). Similar to Zhang et al., we augment encoder-decoder NMT paradigm by introducing a continuous model features process. extend this with co-attention mechanism motivated Parikh al. inference network. Compared vision domain, face additional challenges due discrete nature language, namely posterior collapse. experiment different approaches mitigate issue. show that our...
As more and data is collected for various reasons, the sharing of such becomes paramount to increasing its value. Many applications ranging from smart cities personalized health care require individuals organizations share at an unprecedented scale. Data crucial in today's world, but due privacy security concerns regulation issues, conditions under which occurs needs be carefully specified. Currently, this process done by lawyers requires costly signing legal agreements. In many cases, these...
As more and data is collected for various reasons, the sharing of such becomes paramount to increasing its value. Many applications ranging from smart cities personalized health care require individuals organizations share at an unprecedented scale. Data crucial in today's world, but due privacy security concerns regulation issues, conditions under which occurs needs be carefully specified. Currently, this process done by lawyers requires costly signing legal agreements. In many cases, these...
Automated program repair (APR) aims to help developers improve software reliability by generating patches for buggy programs. Although many code language models (CLM) are developed and effective in tasks such as completion, there has been little comprehensive, in-depth work evaluate CLMs' fixing capabilities fine-tune CLMs the APR task. Firstly, this is first ten on four benchmarks, which shows that surprisingly, best CLM, is, fixes 72% more bugs than state-of-the-art deep-learning...
Purpose Every year, millions of consumers around the world become victims credit card fraud. These individuals have to appeal their companies reverse unauthorized charges. This study aims profile American consumers’ experience when complaints about charges fail produce a resolution. Using large database consumer complaint filings with Consumer Financial Protection Bureau (CFPB), characteristics these are identified, and drivers financial hardship resulting from fraud determined....
Zero forcing is a coloring game played on graph where each vertex initially colored blue or white and the goal to color all vertices by repeated use of (deterministic) change rule starting with as few possible. Probabilistic zero yields discrete dynamical system governed Markov chain. Since in connected any one can eventually entire using probabilistic forcing, expected time do this studied. Given transition matrix for process, an exact formula established propagation time. chains are...
Complex legal agreements enable many real-world applications, from data sharing systems to financial transactions. However, expenses scale with complexity because of the manual processes draft, revise, and enforce agreements. To reduce such costs, we propose a new framework for lawyers develop machine readable agreements, which are automatically verified deployed on Ethereum blockchain. Specifically, our introduces domain specific repositories store human language, web interface Python API...
Numerous Convolutional Neural Networks emerged in the past decade, each varies accuracy, speed, and architecture. From AlexNet to ResNet, CNN models have been developing rapidly, architecture of become more complicated. These are known for their accuracy on ImageNet, so topic this research is explore how can perform differently Fruit 360 dataset. A model constructed specifically three significant developed decade applied dataset result comparison: VGG-16, ResNet-50, MobileNet, SC-3. The...
Download This Paper Open PDF in Browser Add to My Library Share: Permalink Using these links will ensure access this page indefinitely Copy URL DOI
As Internet usage has heavily increased within recent years, money launderers have started to take advantage of Online Financial Transaction (OFT) services facilitate their laundering activities. However, law enforcement struggled understand and detect OFT that criminals use for laundering. To assist in its efforts identify monitor services, we designed the Services Identification Tool (OFTSIT), which crawls determines probability they are services. OFTSIT analyzes a website's content...
Abstract Superannuation funds heavily outsource key fund functions to service providers who play a crucial role in superannuation operations and affecting Australians’ retirement savings. We examine the impact of related party provider usage trustee‐director affiliation on investment performance. find that for‐profit significantly underperform when using providers. The underperformance is more severe board controlled by affiliated trustee‐directors belongs vertically integrated conglomerate...
For the video surveillance system nowadays, identifying color of certain footage is paramount. Every time when it comes to a crime scene, police will be able extract useful information from cameras on scene. Among that important information, "color" plays major role and not affected by size, location, or changes in shape an object. Facts affecting accuracy efficiency realtime identification include material parameter camera lens, parameters infrared for hardware part, software algorithms,...
Humans have the remarkable ability to recognize and acquire novel visual concepts in a zero-shot manner. Given high-level, symbolic description of concept terms previously learned their relations, humans can without seeing any examples. Moreover, they new by parsing communicating structures using relations. Endowing these capabilities machines is pivotal improving generalization capability at inference time. In this work, we introduce Zero-shot Concept Recognition Acquisition (ZeroC),...
We introduce MLE-bench, a benchmark for measuring how well AI agents perform at machine learning engineering. To this end, we curate 75 ML engineering-related competitions from Kaggle, creating diverse set of challenging tasks that test real-world engineering skills such as training models, preparing datasets, and running experiments. establish human baselines each competition using Kaggle's publicly available leaderboards. use open-source agent scaffolds to evaluate several frontier...
Abstract Purpose Pre-trained encoder transformer models have extracted information from unstructured clinic note text but require manual annotation for supervised fine-tuning. Large, Generative Pre- trained Transformers (GPTs) may streamline this process. In study, we explore GPTs in zero- and few-shot learning scenarios to analyze clinical health records. Materials Methods We prompt-engineered LLAMA2 13B optimize performance extracting seizure freedom epilepsy notes compared it against...
Neural language models (LMs) can be used to evaluate the truth of factual statements in two ways: they either queried for statement probabilities, or probed internal representations truthfulness. Past work has found that these procedures sometimes disagree, and probes tend more accurate than LM outputs. This led some researchers conclude LMs “lie’ otherwise encode non-cooperative communicative intents. Is this an description today’s LMs, query–probe disagreement arise other ways? We identify...
Erik Rozi, Niveditha Iyer, Gordon Chi, Enok Choe, Kathy J. Lee, Kevin Liu, Patrick Zander Lack, Jillian Tang, Ethan A. Chi. Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021). 2021.