IISc Bangalore · Computer Science
Computer System Optimizations for ML
Prof. Sumit K. Mandal · 60 free video lectures
Increasing complexity of machine learning (ML) algorithms have necessitated the emergence of specialized computer systems. Some ML algorithms are executed at edge, some are executed on the cloud. In this course, we will delve into different computing kernels of both training and inference of ML…
▶ Play on Open Lecture HallCourse page ↗
Lectures
- Introduction to the course
- Neural Networks
- Introduction to Convolutional Neural Network - 1
- Introduction to Convolutional Neural Network - 2
- Convolutional Neural Network-1
- Convolutional Neural Network-2
- Convolutional Neural Network-3
- Canonical CNN Architectures - 1
- Canonical CNN Architectures - 2
- CNN Computational Optimization Techniques
- Dataflow for CNN Execution - 1
- Dataflow for CNN Execution - 2
- CNN Accelerators - 1
- CNN Accelerators 2
- CNN Accelerators 3
- CNN Accelerators 4
- CNN prunning - 1
- CNN pruning - 2
- Parallelism in CNN - 1
- Parallelism in CNN - 2
- Parallelism in CNN - 3
- Parallelism in CNN - 4
- Parallelism in CNN - 5
- LLMs-1
- LLMs-2
- LLMs-3
- LLMs-4
- LLMs-5
- LLMs-6
- LLMs-7
- LLMs-8
- LLMs-9
- LLMs-10
- Distributed Service
- Fault Tolerance
- Fault-Tolerant Training
- Fundamentals of Graph Neural Networks-1
- Fundamentals of Graph Neural Networks-2
- Hardware Acceleration of Graph Neural Networks-1
- Hardware Acceleration of Graph Neural Networks -2
- ISSAC, Pipelayer
- Pipelayer[contd.], Atomlayer, Simulation platform
- Simulation platform, Area-aware optimization, communication-aware optimization
- Communication-Aware optimization, Achieving minimum communication latency
- Achieving minimum communication latency
- Advanced Hardware Architectures for GNN Acceleration
- Communication-Aware Pruning for IMC
- Re-Transformer
- Performance Evaluation of 2.5D IMC-Based AI Accelerators
- Chiplet-Based Heterogeneous Architectures for LLM Acceleration
- Chiplet-Based Heterogeneous Architectures for LLM Acceleration
- Performance Characterization
- Performance Characterization of Deep Learning Workloads - Part 1
- Performance Characterization of Deep Learning Workloads - Part 2
- Performance Characterization of Deep Learning Workloads - Part 3
- Introduction to AMD Neural Processing Unit (NPU)
- Introduction to AMD Neural Processing Unit (NPU)[contd.]
- Acceleration of ML operation with GPUs
- Training Large model with limited GPU memory
- Recap of the course
From NPTEL (IITs and IISc) · CC BY-NC-SA. More from IISc Bangalore.
Open Lecture Hall is free, with no ads and no sign-up. It is not run by or connected to any university; the lectures are the universities' own free recordings, played from where they host them. About & rights