Advances in Efficient Computing and Privacy-Preserving Techniques

The field of computer science is witnessing significant advancements in efficient computing and privacy-preserving techniques. Researchers are exploring innovative methods to reduce computational costs, improve model performance, and safeguard sensitive information. A key direction in this area is the development of model compression techniques, such as quantization and knowledge distillation, which enable the deployment of large models in resource-constrained environments while maintaining their performance. Additionally, there is a growing interest in designing hardware-aware optimizations and novel architectures that can efficiently execute machine learning workloads. Noteworthy papers in this area include: The paper on quantization's impact on privacy risk in large language models for code, which demonstrates a significant reduction in privacy risk and a positive correlation between task performance and privacy risk. The paper on TROOP, which proposes a set of hardware optimizations to achieve at-the-roofline performance for vector processors on low operational intensity workloads, resulting in significant speedups and improved energy efficiency.

Sources

How Quantization Impacts Privacy Risk on LLMs for Code?

DGEMM without FP64 Arithmetic -- using FP64 Emulation and FP8 Tensor Cores with Ozaki Scheme

Resource-Efficient Automatic Software Vulnerability Assessment via Knowledge Distillation and Particle Swarm Optimization

KBest: Efficient Vector Search on Kunpeng CPU

TROOP: At-the-Roofline Performance for Vector Processors on Low Operational Intensity Workloads

Model Compression vs. Adversarial Robustness: An Empirical Study on Language Models for Code

High-Performance and Power-Efficient Emulation of Matrix Multiplication using INT8 Matrix Engines

Built with on top of