Cost-effective On-device Continual Learning over Memory Hierarchy with Miro

Continual learning (CL) trains NN models incrementally from a continuous stream of tasks. To remember previously learned knowledge, prior studies store old samples over a memory hierarchy and replay them when new tasks arrive. Edge devices that adopt CL to preserve data privacy are typically energy-sensitive and thus require high model accuracy while not compromising energy efficiency, i.e., cost-effectiveness. Our work is the first to explore the design space of hierarchical memory replay-based CL to gain insights into achieving cost-effectiveness on edge devices. We present Miro, a novel system runtime that carefully integrates our insights into the CL framework by enabling it to dynamically configure the CL system based on resource states for the best cost-effectiveness. To reach this goal, Miro also performs online profiling on parameters with clear accuracy-energy trade-offs and adapts to optimal values with low overhead. Extensive evaluations show that Miro significantly outperforms baseline systems we build for comparison, consistently achieving higher cost-effectiveness.

THUEE system description for NIST 2020 SRE CTS challenge

This paper presents the system description of the THUEE team for the NIST 2020 Speaker Recognition Evaluation (SRE) conversational telephone speech (CTS) challenge. The subsystems including ResNet74, ResNet152, and RepVGG-B2 are developed as speaker embedding extractors in this evaluation. We used combined AM-Softmax and AAM-Softmax based loss functions, namely CM-Softmax. We adopted a two-staged training strategy to further improve system performance. We fused all individual systems as our final submission. Our approach leads to excellent performance and ranks 1st in the challenge.

3D Single-pixel imaging with active sampling patterns and learning based reconstruction

Single-pixel imaging (SPI) is significant for applications constrained by transmission bandwidth or lighting band, where 3D SPI can be further realized through capturing signals carrying depth. Sampling strategy and reconstruction algorithm are the key issues of SPI. Traditionally, random patterns are often adopted for sampling, but this blindly passive strategy requires a high sampling rate, and even so, it is difficult to develop a reconstruction algorithm that can maintain higher accuracy and robustness. In this paper, an active strategy is proposed to perform sampling with targeted scanning by designed patterns, from which the spatial information can be easily reordered well. Then, deep learning methods are introduced further to achieve 3D reconstruction, and the ability of deep learning to reconstruct desired information under low sampling rates are analyzed. Abundant experiments verify that our method improves the precision of SPI even if the sampling rate is very low, which has the potential to be extended flexibly in similar systems according to practical needs.