Life Sciences and Health Seminar Aleksandr Chebykin, Solon Pissis

Efficient Neural Architecture Search for ensemble learning; Bidirectional String Anchors: A New String Sampling Mechanism

When
2 Nov 2021 from 4 p.m. to 2 Nov 2021 5 p.m. CET (GMT+0100)
Where
online
Web
Add

Join Zoom Meeting
https://cwi-nl.zoom.us/j/85245134530?pwd=Q0U3a1ZpVUowTXEwU2g4SXQ5SEQrQT09

Meeting ID: 852 4513 4530
Passcode: 317913

Title:      Efficient Neural Architecture Search for ensemble learning
Speaker:    Aleksandr Chebykin
Abstract:   Neural Architecture Search is the problem of learning a task-specific neural network structure from data. During the search, many candidate architectures need to be considered, but it is computationally infeasible to independently train them all. Weight sharing makes the search more efficient but also reduces the space of possible architectures. The tradeoffs of differently restricted search spaces are not obvious. Our idea is to traverse several spaces simultaneously in order to efficiently find an ensemble that is both small and well-performing.

Title:      Bidirectional String Anchors: A New String Sampling Mechanism
Speaker:    Solon Pissis
Abstract:   In a previous LSH seminar, we presented bidirectional string anchors, a new mechanism for sampling strings, which has small density and is computable in linear time. Furthermore, we showed that by using this sampling, we can efficiently construct a small sketch that answers pattern matching queries in near-optimal time. In this seminar, we will present an experimental evaluation, which highlights the applicability of our sketch.