resource-constrained-ml

This repository provides code for machine learning algorithms for edge devices developed at Microsoft Research India.

Code for Stress and Affect Detection on Resource-Constrained Devices

Code for Optimized Arrhythmia Detection on Ultra-Edge Devices

Python implementation of ELM - with optimized speed on MKL-based platforms; Described in conference paper: Radu Dogaru, Ioana Dogaru, "Optimization of extreme learning machines for big data applications using Python", COMM-2018; Allows quantization of weight parameters in both layers and introduces a new and very effective hidden layer nonlinear…

Federated Split Learning via Smashed Activation Gradient Estimation

Embedded Vision for MVS in IoT

This repository is devoted to the development of the facial emotion recognition (FER) system as a final bachelor project at the TU/e. Realised by Blazej Manczak. Supervisors: Dr. Laura Astola (Accenture) and Dr. Vlado Menkovski (TU/e)

subMFL: Compatible subModel Generation for Federated Learning in Device Heterogeneous Environment

A proof of concept implementation of a Data Aware Neural Architecture Search.

Architecture-level study of how early CNN downsampling choices affect bias, generalization, and CPU inference behavior under constrained settings.

Ensemble Deep Random Vector Functional Link with Skip Connections (edRVFL-SC) No GPU required • 100× faster training

ASUTFed: Absolute-Threshold Sparsification with Update Tracking for Communication-Efficient Federated Learning on Resource-Constrained Devices

Image classification using compressed deep neural network ported on resource-constrained platforms.

Theoretical framework and research agenda for culturally attuned and resource-aware foundation models for East African agriculture. Deep Learning Indaba 2025, PMLR 2026.

Knowledge distillation and enhanced subdomain adaptation using graph convolutional network for resource-constrained fault diagnosis — Knowledge-Based Systems (2025) 114251.

Streaming Evaluation Protocol (Prequential + Resource Footprint + Calibration Drift)

Models and their evaluation for paper: Radu Dogaru and Ioana Dogaru "RD-CNN: A Compact and Efficient Convolutional Neural Net for Sound Classification ", ISETC-2020

A Python implementation of the algorithm described in paper Radu Dogaru, Ioana Dogaru, "Optimized Super Fast Support Vector Classifiers Using Python and Acceleration of RBF Computations", (2018) ; There is no output layer learning only a relatively fast selection of support vectors in a RBF-layer optimized for speed. Faster than SVM