Building deep retrieval models in practical applications


Khoa D. Doan

Baidu Research
Fri, Apr 1 2022 - 10:00 am (GMT + 7)
About Speaker

Dr. Khoa D. Doan is currently an AI Researcher at the Cognitive Computing Lab, Baidu Research, USA. He received his PhD in Computer Science at Virginia Tech, and MS in Computer Science at the University of Maryland, College Park. His research focuses on developing practical Deep Learning-To-Hash models and Generative-based Machine Learning approaches in various areas such as Retrieval and AI Security. His work has been published in several Data Mining, Information Retrieval, and Machine Learning conferences. In the past, he spent several years in the industry, working as a Senior Software Engineer, and as a Senior Data Scientist/Engineer at NASA and various advertising companies such as Criteo.


The rapid growth of digital data, especially visual and textual contents, brings many challenges to the problem of finding similar data. Approximate nearest neighbor search methods, such as hashing methods, provide principled approaches that balance the trade-offs between the quality of the guesses and the computational cost for web-scale databases. In this era of data explosion, it is crucial for the retrieval methods to be efficiently trained and for the learned retrieval models to generalize well with limited labeled data, and to be robust to various scenarios such as when the application has noisy data or data that slightly changes over time (i.e., out-of-distribution).

In this talk, he will discuss three lines of work to make hashing-based retrieval methods more suitable for practical uses, to build interpretable retrieval models, and to improve the computational efficiency of real-time ranking with hashing. Specifically, he will introduce a suite of approaches to replace the many quantization losses (>3-4) in existing hashing methods with a single divergence minimization loss, significantly reducing the training time requirement. These approaches have been successfully used to improve both the training process and retrieval performance of the hashing-based methods in several application domains, including computational advertising, and text/image retrieval. Next, he will introduce energy-based generative hashing networks: models that learn the hash functions in energy-based frameworks to improve generalization and robustness. Then, he will introduce an approach that explains the retrieval decision in the graph retrieval domain. Finally, he will introduce one of the first works to use hashing for real-time ranking where the ranking function can be metric (e.g., cosine distance) or non-metric (e.g., neural networks).
Together, these approaches take significant steps forward for building retrieval models that are suitable for practical uses.

Related seminars

Trieu Trinh

Google Deepmind

AlphaGeometry: Solving IMO Geometry without Human Demonstrations
Fri, Jul 5 2024 - 10:00 am (GMT + 7)

Tat-Jun (TJ) Chin

Adelaide University

Quantum Computing in Computer Vision: A Case Study in Robust Geometric Optimisation
Fri, Jun 7 2024 - 11:00 am (GMT + 7)

Fernando De la Torre

Carnegie Mellon University

Human Sensing for AR/VR
Wed, Apr 24 2024 - 07:00 am (GMT + 7)

Anh Nguyen

Microsoft GenAI

The Revolution of Small Language Models
Fri, Mar 8 2024 - 02:30 pm (GMT + 7)