Show Detail »
Schedule Mon Tue Wed Thu
Timezone:
America/Los_Angeles
Filter Events:  
Filter Rooms:  

 

SUN 28 AUG
4 p.m.
Registration Desk:
Registration Check-in Desk
(ends 6:30 PM)
MON 29 AUG
7 a.m.
Registration Desk:
Registration Check-in Desk
(ends 6:00 PM)
8 a.m.
Coffee Break
8:30 a.m.
Opening Remarks
8:45 a.m.
Outstanding Papers [8:45-9:15]
Oral s 8:45-10:15
[8:45] Pathways: Asynchronous Distributed Dataflow for ML
[9:03] QuadraLib: A Performant Quadratic Neural Network Library for Architecture Optimization and Design Exploration
[9:21] Random Offset Block Embedding (ROBE) for compressed embedding tables in deep learning recommendation systems
[9:39] ML-EXray: Visibility into ML Deployment on the Edge
[9:57] GPU Semiring Primitives for Sparse Neighborhood Methods
(ends 9:15 AM)
10:15 a.m.
Break
10:30 a.m.
Invited Talk:
Systems for ML and ML for Systems: A Virtuous Cycle
Kunle Olukotun
(ends 11:30 AM)
11:30 a.m.
Panel:
The Future of MLSys: Industry and Academia - panel
(ends 1:00 PM)
Lunch Break (Box Lunch)
1 p.m.
Distributed and Parallel Learning [1:00-2:15]
Oral s 1:00-2:12
[1:00] BNS-GCN: Efficient Full-Graph Training of Graph Convolutional Networks with Partition-Parallelism and Random Boundary Node Sampling
[1:18] Sequential Aggregation and Rematerialization: Distributed Full-batch Training of Graph Neural Networks on Large Graphs
[1:36] PAPAYA: Practical, Private, and Scalable Federated Learning
[1:54] LightSecAgg: a Lightweight and Versatile Design for Secure Aggregation in Federated Learning
(ends 2:15 PM)
Tutorial:
Gemmini: Enabling Systematic Deep-Learning Architecture Evaluation via Full-Stack Integration
(ends 5:00 PM)
Tutorial:
Time Series Anomaly Detection : Tools, Techniques & Tricks
(ends 5:00 PM)
2:15 p.m.
ML Compilers & Runtime [2:15-3:30]
Oral s 2:15-3:27
[2:15] The CoRa Tensor Compiler: Compilation for Ragged Tensors with Minimal Padding
[2:33] Apollo: Automatic Partition-based Operator Fusion through Layer by Layer Optimization
[2:51] DietCode: Automatic Optimization for Dynamic Tensor Programs
[3:09] Bit-serial Weight Pools: Compression and Arbitrary Precision Execution of Neural Networks on Resource Constrained Processors
(ends 3:30 PM)
3:30 p.m.
Coffee Break
4 p.m.
Systems for ML 1 [4:00-5:30]
Oral s 4:00-5:30
[4:00] MLPerf Mobile Inference Benchmark: An Industry-Standard Open-Source Machine Learning Benchmark for On-Device AI
[4:18] TorchSparse: Efficient Point Cloud Inference Engine
[4:36] Hydrozoa: Dynamic Hybrid-Parallel DNN Training on Serverless Containers
[4:54] Plumber: Diagnosing and Removing Performance Bottlenecks in Machine Learning Data Pipelines
[5:12] Graphiler: Optimizing Graph Neural Networks with Message Passing Data Flow Graph
(ends 5:30 PM)
5:30 p.m.
Reception:
Welcome Reception
(ends 8:00 PM)
TUE 30 AUG
6:30 a.m.
Registration Desk:
Registration Check-in Desk
(ends 5:00 PM)
7:30 a.m.
Coffee Break
8 a.m.
Tutorial:
ML-based Computer System Telemetry Analytics
(ends 10:15 AM)
8:45 a.m.
Systems for ML 2 [8:45-10:15]
Oral s 8:45-10:15
[8:45] QuClassi: A Hybrid Deep Neural Network Architecture based on Quantum State Fidelity
[9:03] VirtualFlow: Decoupling Deep Learning Models from the Underlying Hardware
[9:21] TAGLETS: A System for Automatic Semi-Supervised Learning with Auxiliary Data
[9:39] mmSampler: Efficient Frame Sampler for Multimodal Video Retrieval
[9:57] Sustainable AI: Environmental Implications, Challenges and Opportunities
(ends 10:15 AM)
10:15 a.m.
Break
10:30 a.m.
Invited Talk:
Towards Building a Responsible Data Economy
Dawn Song
(ends 12:00 PM)
11:30 a.m.
Round Table Discussion:
Round Table Discussion
(ends 1:00 PM)
Lunch Break (Box Lunch)
1 p.m.
Tutorial:
Sparsity in ML: Understanding and Optimizing Sparsity in Neural Networks Running on Heterogeneous Systems
(ends 5:00 PM)
ML for Systems [1:00-2:15]
Oral s 1:00-2:12
[1:00] Matchmaker: Data Drift Mitigation in Machine Learning for Large-Scale Systems
[1:18] A Transferable Approach for Partitioning Machine Learning Models on Multi-Chip-Modules
[1:36] SLA-Driven ML INFERENCE FRAMEWORK FOR CLOUDS WITH HETEROGENEOUS ACCELERATORS
[1:54] NURD: Negative-Unlabeled Learning for Online Datacenter Straggler Prediction
(ends 2:15 PM)
2:15 p.m.
Hardware Efficient ML [2:15-3:30]
Oral s 2:15-3:27
[2:15] Collapsible Linear Blocks for Super-Efficient Super Resolution
[2:33] Towards the Co-design of Neural Networks and Accelerators
[2:51] On the Utility of Gradient Compression in Distributed Training Systems
[3:09] Efficient Strong Scaling Through Burst Parallel Training
(ends 3:30 PM)
3:30 p.m.
Coffee Break
4 p.m.
Testing, Debugging and Monitoring & Security [4:00-5:30]
Oral s 4:00-5:30
[4:00] Revelio: ML-Generated Debugging Queries for Finding Root Causes in Distributed Systems
[4:18] Randomness in Neural Network Training: Characterizing the Impact of Tooling
[4:36] Synthesizing Optimal Parallelism Placement and Reduction Strategies on Hierarchical Systems for Deep Learning
[4:54] dPRO: A Generic Performance Diagnosis and Optimization Toolkit for Expediting Distributed DNN Training
[5:12] A Tale of Two Models: Constructing Evasive Attacks on Edge Models
(ends 5:30 PM)
WED 31 AUG
6:30 a.m.
Registration Desk:
Registration Check-in Desk
(ends 5:00 PM)
7:30 a.m.
Coffee Break
8 a.m.
Tutorial:
Online Experimentation for Cloud Applications
(ends 10:15 AM)
8:45 a.m.
Efficient Training [8:45-10:15]
Oral s 8:45-10:15
[8:45] REX: Revisiting Budgeted Training with an Improved Schedule
[9:03] SRIFTY: Swift and Thrifty Distributed Neural Network Training on the Cloud
[9:21] Accelerating Training and Inference of Graph Neural Networks with Fast Sampling and Pipelining
[9:39] Improving Model Training with Multi-fidelity Hyperparameter Evaluation
[9:57] Gyro Dropout: Maximizing Ensemble Effect in Neural Network Training
(ends 10:15 AM)
10:15 a.m.
Break
10:30 a.m.
Invited Talk:
Accelerating Engineering with Machine Learning
Ryan Adams
(ends 12:00 PM)
11:30 a.m.
Lunch Break (Not Provided)
1 p.m.
Tutorial:
ASTRA-sim: Enabling SW/HW Co-Design Exploration for Distributed Deep Learning Training Platforms
(ends 5:00 PM)
ML Programming Models and Abstractions & Interpretability and Explainability of ML [1:00-2:15]
Oral s 1:00-2:12
[1:00] Understanding GNN Computational Graph: A Coordinated Computation, IO, and Memory Perspective
[1:18] torch.fx: Practical Program Capture and Transformation for Deep Learning in Python
[1:36] FROTE: Feedback Rule-Driven Oversampling for Editing Models
[1:54] TyXe: Pyro-based Bayesian neural nets for Pytorch
(ends 2:15 PM)
Tutorial:
Training-Free Approaches for Edge AI: Challenges, Opportunities and Progress
(ends 5:00 PM)
2:15 p.m.
Efficient Inference and Model Serving [2:15-4:03]
Oral s 2:15-4:03
[2:15] ULPPACK: Fast Sub-8-bit Matrix Multiply on Commodity SIMD Hardware
[2:33] AccMPEG: Optimizing Video Encoding for Accurate Video Analytics
[2:51] HALOS: Hashing Large Output Space for Cheap Inference
[3:09] Learning Compressed Embeddings for On-Device Inference
[3:27] Bolt: Bridging the Gap between Auto-tuners and Hardware-native Performance
[3:45] URSABench: A System for Comprehensive Benchmarking of Bayesian Deep Neural Network Models and Inference methods
(ends 4:03 PM)
5 p.m.
Closing Remarks
THU 1 SEP
7 a.m.
Registration Desk:
Registration Check-in Desk
(ends 12:00 PM)
8 a.m.
Workshop:
Practical Adoption Challenges of ML for Systems in Industry (PACMI)
(ends 5:00 PM)
Workshop:
Benchmarking Machine Learning Workloads on Emerging Hardware
(ends 5:00 PM)
Symposium:
Chips & Compilers
(ends 5:00 PM)
Workshop:
Cloud Intelligence: AI/ML for Efficient and Manageable Cloud Services
(ends 5:00 PM)
Workshop:
Cross-Community Federated Learning: Algorithms, Systems and Co-designs
(ends 5:00 PM)
10 a.m.
Coffee Break AM
3 p.m.
Coffee Break PM