Search:
Match:
4 results
Research Paper#Computer Vision, Deep Learning, Model Compression, Robustness🔬 ResearchAnalyzed: Jan 3, 2026 06:17

Compression Techniques and CNN Robustness

Published:Dec 31, 2025 17:00
1 min read
ArXiv

Analysis

This paper addresses a critical practical concern: the impact of model compression, essential for resource-constrained devices, on the robustness of CNNs against real-world corruptions. The study's focus on quantization, pruning, and weight clustering, combined with a multi-objective assessment, provides valuable insights for practitioners deploying computer vision systems. The use of CIFAR-10-C and CIFAR-100-C datasets for evaluation adds to the paper's practical relevance.
Reference

“Certain compression strategies not only preserve but can also improve robustness, particularly on networks with more complex architectures.”

PermalinkArXiv
Research Paper#Computer Vision, Deep Learning, Image Classification🔬 ResearchAnalyzed: Jan 3, 2026 15:53

Bayesian Self-Distillation Improves Image Classification

Published:Dec 30, 2025 11:48
1 min read
ArXiv

Analysis

This paper introduces Bayesian Self-Distillation (BSD), a novel approach to training deep neural networks for image classification. It addresses the limitations of traditional supervised learning and existing self-distillation methods by using Bayesian inference to create sample-specific target distributions. The key advantage is that BSD avoids reliance on hard targets after initialization, leading to improved accuracy, calibration, robustness, and performance under label noise. The results demonstrate significant improvements over existing methods across various architectures and datasets.
Reference

“BSD consistently yields higher test accuracy (e.g. +1.4% for ResNet-50 on CIFAR-100) and significantly lower Expected Calibration Error (ECE) (-40% ResNet-50, CIFAR-100) than existing architecture-preserving self-distillation methods.”

PermalinkArXiv
Research Paper#Federated Learning, Edge Computing, Deep Learning🔬 ResearchAnalyzed: Jan 3, 2026 19:06

Energy and Memory-Efficient Federated Learning with Ordered Layer Freezing

Published:Dec 29, 2025 04:39
1 min read
ArXiv

Analysis

This paper addresses the challenges of Federated Learning (FL) on resource-constrained edge devices in the IoT. It proposes a novel approach, FedOLF, that improves efficiency by freezing layers in a predefined order, reducing computation and memory requirements. The incorporation of Tensor Operation Approximation (TOA) further enhances energy efficiency and reduces communication costs. The paper's significance lies in its potential to enable more practical and scalable FL deployments on edge devices.
Reference

“FedOLF achieves at least 0.3%, 6.4%, 5.81%, 4.4%, 6.27% and 1.29% higher accuracy than existing works respectively on EMNIST (with CNN), CIFAR-10 (with AlexNet), CIFAR-100 (with ResNet20 and ResNet44), and CINIC-10 (with ResNet20 and ResNet44), along with higher energy efficiency and lower memory footprint.”

PermalinkArXiv
Research Paper#Large Language Models (LLMs), Machine Learning, Multi-Expert Systems🔬 ResearchAnalyzed: Jan 3, 2026 19:28

Learning with Multi-Expert Deferral for LLMs

Published:Dec 28, 2025 11:33
1 min read
ArXiv

Analysis

This paper addresses critical challenges of Large Language Models (LLMs) such as hallucinations and high inference costs. It proposes a framework for learning with multi-expert deferral, where uncertain inputs are routed to more capable experts and simpler queries to smaller models. This approach aims to improve reliability and efficiency. The paper provides theoretical guarantees and introduces new algorithms with empirical validation on benchmark datasets.
Reference

“The paper introduces new surrogate losses and proves strong non-asymptotic, hypothesis set-specific consistency guarantees, resolving existing open questions.”

PermalinkArXiv