Welcome to the graduate course on Systems and Machine Learning! This is a research-oriented course covering topics on Systems for Machine Learning and Machine Learning for Systems.

Machine learning is transforming several domains ranging from natural language processing to drug discovery today. One of the key factors that enabled rapid progress in ML/AI in recent years has been fast-evolving underlying hardware and software platforms. In this course, we will cover recent advancements in research and industry on machine learning systems that enabled the AI/ML revolution. Specific topics include domain-specific architectures, deep learning frameworks and compilers, networking and scheduling in deep learning clusters, etc. We will also discuss practical challenges in deploying such systems. In the second half of the course, we will explore how machine learning has been employed to tackle various networking and systems challenges such as Internet congestion control, adaptive bitrate selection in video streaming, flow prediction, etc.

Instructor: Sangeetha Abdu Jyothi

Class Hours: MW 5:00-6:20 PM

Location: PCB 1200 and live stream: https://uci.zoom.us/j/92825560246?pwd=TW9mQjVjV01xK3VQWDc3SVZrU0pYQT09 (requires Zoom log in with UCI account)

Office Hours: Wed 1-2 pm (https://uci.zoom.us/j/92149956397?pwd=RStDV2J1MmpNR0RJc0xabEo2bFRmdz09) and MW 6:20-6:50 pm (in person after class)

TA Office Hours: Mon 2:30 - 4:00 PM, DBH 3243

Course Policies: Course Policies

Prerequisites: Understanding of basic concepts in machine learning and systems (taken at least one undergrad course in ML and (networking or distributed systems))

Schedule: (more optional papers will be added as the quarter progresses)

Date	Category	Lecture	Required Reading	Optional
9/26/2022		Lec 1: Introduction
9/28/2022	Systems For ML	Lec 2: Domain Specific Architectures	In-Datacenter Performance Analysis of a Tensor Processing Unit (ISCA'17)	A Configurable Cloud-Scale DNN Processor for Real-Time AI (ISCA'18)
10/3/2022	Systems For ML	Lec 3: DL Frameworks	PyTorch (NeurIPS'19)	TensorFlow (OSDI'16) MXNet CNTK (KDD'16)
10/5/2022	Systems For ML	Lec 4: DL compilers	TVM (OSDI'18)	Glow nGraph XLA TensorComprehensions
10/10/2022	Systems For ML	Lec 5: Networking Challenges in DNN training	Parameter Server (OSDI'14)	Horovod (arXiv'18) BytePS (OSDI'20) TicTac (MLSys'19) P3 (MLSys'19)
10/12/2022	Systems For ML	Lec 6: Cluster scheduling for DL workloads	Gavel (OSDI'20)	AntMan (OSDI'20) Themis (NSDI'20) Gandiva (OSDI'18)
10/17/2022	Systems For ML	Lec 7: Automated ML: Hyperparameter Tuning and Neural Architectural Search (NAS)	Hyperband (JMLR'18)	Cerebro (VLDB'20) Vizier (KDD'17) ASHA (MLSys'20) Designing Neural Architectures using RL (ICLR'17)
10/19/2022	Systems For ML	Lec 8: Scaling Gradient Boosting and RL	Asynchronous Methods for Deep Reinforcement Learning (ICML'18)	RLlib (ICML'18) XGBoost (KDD'16)
10/24/2022	Systems For ML	Lec 9: ML at the Edge	Towards Federated Learning at Scale: System Design (MLSys'19)
10/26/2022	Systems For ML	Lec 10: Challenges in Operational ML Systems	Hidden Technical Debt in Machine Learning Systems (NeurIPS'15)	TFX (KDD'17) Applied Machine Learning at Facebook: A Datacenter Infrastructure Perspective (HPCA'18)
10/31/2022	ML for Systems	Lec 11: Deep RL and Challenges in the real world	Deep RL tutorial (background reading, not for review) Challenges of Real-World Reinforcement Learning (ICML'19)
11/2/2022	ML for Systems	Lec 12: Deep RL for Congestion Control	Aurora (ICML'19)
11/7/2022	ML for Systems	Lec 13: Deep RL for Video Streaming (ABR control)	Pensieve (SIGCOMM'17)
11/9/2022	ML for Systems	Lec 14: Classic control + Learning (congestion control & ABR)	Classic Meets Modern (SIGCOMM'20)	Learning in-situ (ATC'20)
11/14/2022	ML for Systems	Lec 15: ML for Index Structures	Learned Index Structures (SIGMOD'18)
11/16/2022	ML for Systems	Lec 16: ML for Flow Prediction	Flux (NSDI'19)
11/21/2022	ML for Systems	Lec 17: Interpretability of RL-based controllers	Metis (SIGCOMM'20)
11/23/2022		Final Project Presentations
11/28/2022		Final Project Presentations
11/30/2022		Final Project Presentations

Grading

Paper Summaries: 30%

Class participation: 20%

Final Project: 50%

(Title and plan: 5%

Checkpoint 1: 10%

Checkpoint 2 Report: 10%

In-class Presentation: 15%

Final report: 10%)