Explore a neural network framework utilizing emerging technologies for screening Diabetic Retinopathy in this 55-minute tinyML Talk by Koteswararao Chilakala, a Masters graduate in Embedded Systems from TU Delft. Delve into the challenges of addressing Diabetic Retinopathy, a leading cause of permanent vision loss affecting millions globally, particularly in remote and low-income settings. Learn about advancements in imaging technologies enabling fundus examinations on handheld devices and the development of an integrated solution delivering high compute at ultra-low power consumption. Discover the process of creating robust CNN models through vigilant training on diverse datasets, a new binary labeling scheme maximizing softmax probabilities, and implementation on resistive random access memory (RRAM) based computational in memory (CIM) architecture. Gain insights into significant improvements in energy consumption and latency compared to traditional computing platforms. The talk covers various aspects including data processing, model training, compression techniques, performance metrics, and potential high-value or safety-critical use cases in AI-powered solutions across industries.
TinyML Talks Germany - Neural Network Framework Using Emerging Technologies for Screening Diabetic
Overview
Syllabus
Intro
Diabetic Retinopathy(DR)
Convolutional Neural Networks CN
Related Work - CNN Literature
Computation in memory (CIM) Architecture
Resistive Random Access Memory RE
Tasks in this work
Performance Metrics
Data Processing
Model Training
Severity Label (SL)
Model Compression
Multi-class accuracy
Severity Label Accuracy
Model Evaluation
Compression Schemes
Latency
Energy Consumption - Quantization
Conclusion
High-Value or Safety-Critical Use Cases? For your most important projects, use
Renesas is enabling the next generation of Al-powered solutions that will revolutionise every industry sector
Silver Strategic Partners
Taught by
tinyML
