Edge AI Research Areas

Lightweight On-Device Intelligence

Develops compact models and adaptive execution methods for real-time intelligence directly on edge devices, reducing memory, computation, communication, and energy requirements.

Paper 1: GestLite: Lightweight Gesture Detection for Edge AI via Event-Driven Model and Input Adaptations

Paper 2: Scale-Gest: Scalable Model-Space Synthesis and Runtime Selection for On-Device Gesture Detection

Paper 3: tinyDigiClones: A Multi-Modal LLM-Based Framework for Edge-Optimized Personalized Avatars

Embedded Transformer Acceleration

Explores architectures, simulation tools, memory optimizations, and accelerator designs for executing Transformer models efficiently under embedded latency, energy, and area constraints.

Paper 1: TRAPTI: Time-Resolved Analysis for SRAM Banking and Power Gating Optimization in Embedded Transformer Inference

Paper 2: TransInferSim: Towards Fast and Accurate Evaluation of Embedded Hardware Accelerators for Transformer Networks

Paper 3: SwiftTron: An Efficient Hardware Accelerator for Quantized Transformers

Memory & Energy-Efficient Inference

Develops data-mapping, quantization, low-precision processing, and memory-management techniques that reduce data movement and improve the performance and energy efficiency of embedded inference.

Paper 1: PENDRAM: Enabling High-Performance and Energy-Efficient Processing of Deep Neural Networks through a Generalized DRAM Data Mapping Policy

Paper 2: DRMap: A Generic DRAM Data Mapping Policy for Energy-Efficient Processing of Convolutional Neural Networks

Paper 3: CoNLoCNN: Exploiting Correlation and Non-Uniform Quantization for Energy-Efficient Low-precision Deep Convolutional Neural Networks

Hardware/Software Co-Design

Jointly optimizes neural models, software stacks, memory systems, and hardware accelerators to meet target accuracy, latency, energy, and platform-resource constraints.

Paper 1: CuriousRL: Curiosity-Driven Reinforcement Learning for Adaptive Locomotion in Quadruped Robots

Paper 2: Continual Learning for Real-World Autonomous Systems: Algorithms, Challenges and Frameworks

Paper 3: ESM: A Framework for Building Effective Surrogate Models for Hardware-Aware Neural Architecture Search

Approximate Computing for Edge Acceleration

Investigates controlled approximation and configurable-precision computation to reduce the cost of edge-AI workloads while managing accuracy, efficiency, and hardware-complexity trade-offs.

Paper 1: Enabling Capsule Networks at the Edge through Approximate Softmax and Squash Operations

Paper 2: Approximate Acceleration for CNN-based Applications on IoT Edge Devices

Paper 3: Approximate Computing Survey, Part II: Application-Specific & Architectural Approximation Techniques and Applications

Robust & Secure Edge AI

Investigates controlled approximation and configurable-precision computation to reduce the cost of edge-AI workloads while managing accuracy, efficiency, and hardware-complexity trade-offs.

Paper 1: PatchBlock: A Lightweight Defense Against Adversarial Patches for Embedded EdgeAI Devices

Paper 2: RoHNAS: A Neural Architecture Search Framework With Conjoint Optimization for Adversarial Robustness and Hardware Efficiency of Convolutional and Capsule Networks

Paper 3: Towards Energy-Efficient and Secure Edge AI: A Cross-Layer Framework