Modular and Adaptive Computational Frameworks

The recent developments in the research area have shown a strong focus on enhancing the efficiency and flexibility of various computational frameworks and architectures. There is a notable trend towards modular and open-source solutions that facilitate rapid prototyping and customization, particularly in the fields of deep learning, network interface card (NIC) design, and optimization algorithms. These advancements aim to address the complexities and resource constraints inherent in modern computational tasks, such as those involving deep neural networks (DNNs) and real-time optimal control in robotics. Additionally, there is a growing interest in dynamic and adaptive learning strategies, such as those leveraging bandit algorithms for hyperparameter tuning in deep reinforcement learning, which promise to improve model performance and convergence rates. The integration of high-level synthesis with physical layout optimization in FPGA design is also emerging as a critical area, aiming to balance performance and design productivity. Overall, the field is moving towards more adaptable, efficient, and scalable solutions that can be tailored to specific application requirements.

Noteworthy papers include:

  • A framework for enabling algorithmic design choice exploration in DNNs, which offers fine-grain control and high-performance implementations.
  • A dynamic learning rate approach for deep reinforcement learning, utilizing a bandit algorithm to adaptively select optimal learning rates.

Sources

A Framework to Enable Algorithmic Design Choice Exploration in DNNs

RISC-V V Vector Extension (RVV) with reduced number of vector registers

JingZhao: A Framework for Rapid NIC Prototyping in the Domain-Specific-Network Era

Efficient Hyperparameter Importance Assessment for CNNs

The State of Julia for Scientific Machine Learning

Technical Report of 1:10 Scale Autonomous Vehicle Robot

Comparative Performance of Collaborative Bandit Algorithms: Effect of Sparsity and Exploration Intensity

Accelerating Python Applications with Dask and ProxyStore

Design Space Exploration of Embedded SoC Architectures for Real-Time Optimal Control

Approaching Metaheuristic Deep Learning Combos for Automated Data Mining

Dynamic Learning Rate for Deep Reinforcement Learning: A Bandit Approach

modOpt: A modular development environment and library for optimization algorithms

ORANSlice: An Open-Source 5G Network Slicing Platform for O-RAN

RapidStream IR: Infrastructure for FPGA High-Level Physical Synthesis

Transformers4NewsRec: A Transformer-based News Recommendation Framework

A Self-Constructing Multi-Expert Fuzzy System for High-dimensional Data Classification

AutoAL: Automated Active Learning with Differentiable Query Strategy Search

Built with on top of