A5003172820- Advanced Image and Video Retrieval Techniques
- Generative Adversarial Networks and Image Synthesis
- Text and Document Classification Technologies
- Video Analysis and Summarization
- Multimodal Machine Learning Applications
- Topic Modeling
- Web Data Mining and Analysis
- Image Retrieval and Classification Techniques
- Natural Language Processing Techniques
- Information Retrieval and Search Behavior
- Face recognition and analysis
- Constraint Satisfaction and Optimization
- Sentiment Analysis and Opinion Mining
- Software Engineering Research
- Caching and Content Delivery
- Handwritten Text Recognition Techniques
Amazon (United States)
2023
Santa Clara University
2017-2022
Walmart (United States)
2022
As the amount of textual data has been rapidly increasing over past decade, efficient similarity search methods have become a crucial component large-scale information retrieval systems. A popular strategy is to represent original samples by compact binary codes through hashing. spectrum machine learning utilized, but they often lack expressiveness and flexibility in modeling learn effective representations. The recent advances deep wide range applications demonstrated its capability robust...
With an ever increasing amount of data available on the web, fast similarity search has become critical component for large-scale information retrieval systems. One solution is semantic hashing which designs binary codes to accelerate search. Recently, deep learning been successfully applied problem and produces high-quality compact compared traditional methods. However, most state-of-the-art approaches require large amounts hand-labeled training are often expensive time consuming collect....
In product search, the retrieval of candidate products before re-ranking is more mission critical and challenging than other search like web especially for tail queries, which have a complex specific intent. this paper, we present hybrid system e-commerce deployed at Walmart that combines traditional inverted index embedding-based neural to better answer user queries. Our significantly improved relevance engine, measured by both offline online evaluations. The improvements were achieved...
Product ranking is a crucial component for many e-commerce services. One of the major challenges in product search vocabulary mismatch between query and products, which may be larger gap problem compared to other information retrieval domains. While there growing collection neural learning match methods aimed specifically at overcoming this issue, they do not leverage recent advances large language models search. On hand, often deals with multiple types engagement signals such as clicks,...
When shopping for fashion, customers often look products which can complement their current outfit. For example, want to buy a jacket go well with jeans and sneakers. To address the task of fashion matching, we propose neural compatibility model ranking based on matching input The contribution our work is twofold. First, demonstrate that product descriptions contain rich information about comparability has not been fully utilized in prior work. Secondly, exploit such useful from text data by...
As the amount of textual data has been rapidly increasing over past decade, efficient similarity search methods have become a crucial component large-scale information retrieval systems. A popular strategy is to represent original samples by compact binary codes through hashing. spectrum machine learning utilized, but they often lack expressiveness and flexibility in modeling learn effective representations. The recent advances deep wide range applications demonstrated its capability robust...
In product search, the retrieval of candidate products before re-ranking is more critical and challenging than other search like web especially for tail queries, which have a complex specific intent. this paper, we present hybrid system e-commerce deployed at Walmart that combines traditional inverted index embedding-based neural to better answer user queries. Our significantly improved relevance engine, measured by both offline online evaluations. The improvements were achieved through...
Constrained decoding with lookahead heuristics (CDLH) is a highly effective method for aligning LLM generations to human preferences. However, the extensive roll-out operations each generated token makes CDLH prohibitively expensive, resulting in low adoption practice. In contrast, common strategies such as greedy are extremely efficient, but achieve very constraint satisfaction. We propose constrained speculative lookaheads (CDSL), technique that significantly improves upon inference...
Knowledge-intensive programming Q&A is an active research area in industry. Its application boosts developer productivity by aiding developers quickly finding answers from the vast amount of information on Internet. In this study, we propose ProQANS and its variants ReProQANS ReAugProQANS to tackle Q&A. a neural search approach that leverages unlabeled data Internet (such as StackOverflow) mitigate cold-start problem. extends utilizing reformulated queries with novel triplet loss. We further...