Timezone: »

Streaming Active Learning with Deep Neural Networks
Akanksha Saran · Safoora Yousefi · Akshay Krishnamurthy · John Langford · Jordan Ash

Thu Jul 27 04:30 PM -- 06:00 PM (PDT) @ Exhibit Hall 1 #716
Event URL: https://github.com/asaran/VeSSAL »

Active learning is perhaps most naturally posed as an online learning problem. However, prior active learning approaches with deep neural networks assume offline access to the entire dataset ahead of time. This paper proposes VeSSAL, a new algorithm for batch active learning with deep neural networks in streaming settings, which samples groups of points to query for labels at the moment they are encountered. Our approach trades off between uncertainty and diversity of queried samples to match a desired query rate without requiring any hand-tuned hyperparameters. Altogether, we expand the applicability of deep neural networks to realistic active learning scenarios, such as applications relevant to HCI and large, fractured datasets.

Author Information

Akanksha Saran (Microsoft Research)

Akanksha is a Postdoctoral Researcher at Microsoft Research NYC. Her research focuses on modeling human behavior across multimodal interfaces and designing human-interactive machine learning algorithms for sequential decision-making settings such as imitation learning, interaction-grounded learning, and active learning. She is particularly interested in alleviating barriers of access for end-users who can benefit from interactive machine learning algorithms in the real-world. Her interdisciplinary work attempts to (1) further our understanding of how humans interact with artificial learning agents and (2) design efficient algorithms which leverage varied forms of human data for real-world interactive applications. Akanksha earned her PhD from the Department of Computer Science at the University of Texas at Austin in 2021.

Safoora Yousefi (Microsoft)
Akshay Krishnamurthy (Microsoft)
John Langford (MSR)
Jordan Ash (Microsoft Research NYC)

More from the Same Authors