Nam Thoai

Duong H. Le, Trung-Nhan Vo, Nam Thoai

Network pruning is one of the most dominant methods for reducing the heavy inference cost of deep neural networks. Existing methods often iteratively prune networks to attain high compression ratio without incurring significant loss in performance. However, we argue that conventional methods for retraining pruned networks (i.e., using small, fixed learning rate) are inadequate as they completely ignore the benefits from snapshots of iterative pruning. In this work, we show that strong ensembles can be constructed from snapshots of iterative pruning, which achieve competitive performance and vary in network structure. Furthermore, we present simple, general and effective pipeline that generates strong ensembles of networks during pruning with large learning rate restarting, and utilizes knowledge distillation with those ensembles to improve the predictive power of compact models. In standard image classification benchmarks such as CIFAR and Tiny-Imagenet, we advance state-of-the-art pruning ratio of structured pruning by integrating simple l1-norm filters pruning into our pipeline. Specifically, we reduce 75-80% of total parameters and 65-70% MACs of numerous variants of ResNet architectures while having comparable or better performance than that of original networks. Code associate with this paper is made publicly available at https://github.com/lehduong/kesi.

Duc-Phuong Doan-Ngo, Thanh-Dang Diep, Thanh Nguyen-Duc et al.

Recognizability, a key perceptual factor in human face processing, strongly affects the performance of face recognition (FR) systems in both verification and identification tasks. Effectively using recognizability to enhance feature representation remains challenging. In deep FR, the loss function plays a crucial role in shaping how features are embedded. However, current methods have two main drawbacks: (i) recognizability is only partially captured through soft margin constraints, resulting in weaker quality representation and lower discrimination, especially for low-quality or ambiguous faces; (ii) mutual overlapping gradients between feature direction and magnitude introduce undesirable interactions during optimization, causing instability and confusion in hypersphere planning, which may result in poor generalization, and entangled representations where recognizability and identity are not cleanly separated. To address these issues, we introduce a hard margin strategy - Quality Control Face (QCFace), which overcomes the mutual overlapping gradient problem and enables the clear decoupling of recognizability from identity representation. Based on this strategy, a novel hard-margin-based loss function employs a guidance factor for hypersphere planning, simultaneously optimizing for recognition ability and explicit recognizability representation. Extensive experiments confirm that QCFace not only provides robust and quantifiable recognizability encoding but also achieves state-of-the-art performance in both verification and identification benchmarks compared to existing recognizability-based losses.

Nguyen Quang-Hung, Pham Dac Nien, Nguyen Hoai Nam et al.

Energy efficiency has become an important measurement of scheduling algorithm for private cloud. The challenge is trade-off between minimizing of energy consumption and satisfying Quality of Service (QoS) (e.g. performance or resource availability on time for reservation request). We consider resource needs in context of a private cloud system to provide resources for applications in teaching and researching. In which users request computing resources for laboratory classes at start times and non-interrupted duration in some hours in prior. Many previous works are based on migrating techniques to move online virtual machines (VMs) from low utilization hosts and turn these hosts off to reduce energy consumption. However, the techniques for migration of VMs could not use in our case. In this paper, a genetic algorithm for power-aware in scheduling of resource allocation (GAPA) has been proposed to solve the static virtual machine allocation problem (SVMAP). Due to limited resources (i.e. memory) for executing simulation, we created a workload that contains a sample of one-day timetable of lab hours in our university. We evaluate the GAPA and a baseline scheduling algorithm (BFD), which sorts list of virtual machines in start time (i.e. earliest start time first) and using best-fit decreasing (i.e. least increased power consumption) algorithm, for solving the same SVMAP. As a result, the GAPA algorithm obtains total energy consumption is lower than the baseline algorithm on simulated experimentation.