oneAPI DevSummit for AI and HPC 2022

December 6, 2022 | Day One: 9:00 - 5:00 pm CT; Day Two: 9:00 - 3:30 pm CT

The oneAPI DevSummit for AI and HPC is a free virtual community conference for those who are interested in performant, cross-architecture programming and want to evolve heterogeneous computing to its fullest potential.

Industry experts take a deep dive into cross-architecture software development, spanning:
– AI and analytics using industry-standard frameworks and tools, performance-optimized by oneAPI
– What’s new with oneAPI including Intel® AI Analytics and HPC Toolkits
– Overviews of key performance analysis tools
– The latest on the oneAPI specification
– Hands-on workshops with TensorFlow and PyTorch

You’ll get to experience a luminary keynote, tech talks, hands-on tutorials, and a perspective on AI and HPC to catapult you to the next level in your developer journey. This event promises to deliver a top-notch view of the aspects of AI and HPC vital to your development career.

Agenda

Day 1
Day 2

Day 1

9:00 - 9:15 CT

Introduction

Presenting

9:15 - 10:00 CT Keynote

AI Software and Hardware Acceleration

We are all aware of the challenges involved in moving AI from pilot to production due to the diversity of data, models and deployment targets. Learn how Intel’s hardware architectures and oneAPI-powered AI software tools are improving application performance and developer productivity, and how our ecosystem engagements accelerate business results. We will showcase how you can leverage our CPUs, GPUs and Habana processors along with…
Presenting

10:05 - 10:25 CT Tech Talk

Using oneAPI to Predict Anonymous Web Visitor Behavior

Would you like to know which anonymous visitor is likely to spend money on your website? We will show you a production example of how to leverage Intel oneAPI to train and infer website anonymous visitor purchasing behaviors at scale. SiteMana will showcase a production ready demo of visitor purchase intent prediction via the oneAPI Regression Usage Model, using Regression Stump to perform inferencing in real time via daal4py using…
Presenting

10:25 - 10:35 CT Break

10:35 - 11:05 CT Tech Talk

Efficient Inference and Training of Large Neural Network Models

​The memory consumption and computational cost of state-of-the-art deep neural network models are dramatically increasing. Therefore, it is beneficial to apply efficient deep learning to both inference and training. In this talk, we present our progress regarding this topic. First, we introduce LTP, which uses pruning to accelerate inference. Then we talk about staged training for transformers, and TASC, which are designed to accelerate training.…
Presenting

11:10 - 12:10 CT Hands-On Training

Accelerating PyTorch Deep Learning Models on Intel XPUs

​In this session, we will present optimizations for Intel XPUs in PyTorch Upstream, Intel® Extension for PyTorch, and popular projects in the PyTorch ecosystem such as Hugging Face. Through demos, we will share with audiences the best-known methods to make full use of the optimizations to get the best performance in their deployments with Intel products, and we will give our audience opportunities to do…
Presenting

12:10 - 12:40 CT Lunch

12:40 - 1:10 CT Lightning Talk

Hacking the Hackathon using FastAI and IPEX

Your chances of winning a hackathon competition are directly proportional to the number of experiments that you conduct. Learn how we quickly iterated between testing ideas and models, and analyzing errors, to ace the AI for Social Good Hackathon at Intel Innovation 2022. We will show how we used fastai to quickly build a starting point, and how we leveraged Intel’s AI tools and frameworks…
Presenting

1:15 -1:45 CT Tech Talk

Spatial Single Cell Analysis Using oneAPI AI Analytics Toolkit

The topic focuses on implementation of oneAPI AI toolkit in Medical Imaging. We will be exploring single cell data (eg:- RNA sequence) through imaging and gene expression profiling. We will be porting Squidpy to oneAPI AI toolkit.Squidpy produces highly interactive visualizations that enable intuitive exploration of Single Cell spatialmolecular data and has several optional uses (e.g. analyzing neighborhood graphs, and compute spatial statistics ) to…
Presenting

1:45 - 1:55 CT Break

1:55 - 2:10 CT Community Forum

Introducing the oneAPI Community Forum

The oneAPI initiative was announced around 3 years ago to facilitate the publishing of the oneAPI specification, a set of defined programming interfaces for writing software for accelerators such as GPUs. Multiple versions of this specification have been successfully released and regular technical discussions have been facilitated through a set of Technical Advisory Boards. This session will present the next steps for this effort, a…
Presenting

2:10 - 2:40 CT TAB Talk

RISC-V Vectors and oneAPI: Accelerating the Future of Heterogeneous Compute

The partnership of oneAPI and RISC-V advances innovation from open source software down into hardware. Open standards enable us to build heterogeneous compute products where engineers can choose the right tools for the job. The RISC-V community has been working with the oneAPI community to help build an open AI ecosystem. This spans the development stack from tools to compilers, and down to the hardware…
Presenting

2:40 - 3:40 CT Hands-On Training

Leveraging Default Intel Optimizations for TensorFlow

This lab will provide hands-on experience using official TensorFlow for Transfer Learning on a common use case and demonstrate its ease of use. We will run inference on the re-trained model, optimize the model for best latency and finally deploy it using TensorFlow serving on a 3rd Gen Intel® Xeon® Scalable Processors with AI acceleration. We will conclude with a demo highlighting improved AI performance…
Presenting

3:40 - 4:00 CT

Conclusion

Presenting

4:00 - 5:00 CT

Happy Hour

Presenting

Day 2

9:00 - 9:15 CT

Introduction

Presenting

9:15 - 10:00 CT Keynote

Performance Portability: From Fantasy to Reality

As we go through what many in high performance computing are calling a “Cambrian Explosion” of hardware architectures, what we would ideally like to be able to (delete) do is write software once and achieve maximum performance on all these new innovative platforms. We know that this ideal is not possible. However, we are making rapid progress towards achieving a level of performance portability that’s…
Presenting

10:05 - 10:35 CT Tech Talk

Preparing Applications for Aurora: Early Successes in Porting HPC Workloads to PVC

In the coming year, Argonne will begin deploying the Aurora Exascale supercomputer powered by Intel Ponte-Vecchio (PVC) GPUs and Sapphire Rapids CPUs. This talk will provide an overview of the Aurora system hardware and software, and the progress that has been made in preparing complex HPC applications to run on it. A diverse portfolio of applications is being developed to perform novel leading edge science…
Presenting

10:35 - 10:45 CT Break

10:45 - 11:15 CT Tech Talk

High-Performance Neuromorphic Sensor Processing

Neuromorphic, or event-based, sensors capture events at a microsecond resolution, which requires low-latency processing via FPGAs for real-time performance. By leveraging both the temporal and spatial components of the events, some algorithms, like a Hierarchy of event- based Time-Surfaces (HOTS), we find relationships between the two components for enhanced feature extraction and object detection. In contrast, using k-means clustering on spatial information can explore the…
Presenting

11:20 - 11:50 CT Tech Talk

Preparing NAMD for the Aurora Supercomputer

NAMD is a well-known, highly scalable parallel molecular dynamics application. A past recipient of the Gordon Bell Award in 2002 and a Gordon Bell Special Prize in 2020, NAMD is capable of scaling simulations to tens of thousands of CPU cores and thousands of GPUs, providing a tool for biomedical researchers to understand protein assemblies, cell membrane channels, and even entire virus capsids at atomic-scale…
Presenting

11:50 - 12:20 CT Lunch

12:20 - 12:50 CT Tech Talk

Performance Evolution of Different SYCL Implementations on the Basis of PLSSVM

We developed PLSSVM, a GPU accelerated Parallel Least Squares Support Vector Machine, able to classify dense data sets with hundreds of thousand data points and more than thousand features beating the state-of-the-art SMO implementations like LIBSVM. Additionally, we support a plethora of different hardware architectures like any CPU and GPUs from Intel, NVIDIA, and AMD using different backends written in OpenMP, CUDA, HIP, OpenCL, and…
Presenting

12:55 - 2:25 CT Live Tutorial

oneAPI Dev Workflows for SYCL & Fortran: Live Tutorial / Demo Showcase

At SuperComputing ’22 we demonstrated demonstrating key aspects of the developer workflow using Intel® oneAPI tools to program for the Intel® Data Center GPU Max Series, formerly codenamed Ponte Vecchio (PVC), all using cross vendor, cross architecture standards based techniques. We would like to share several use cases with you in the form of a live code tutorial + demo, representing the latest and most…
Presenting

2:25 - 2:35 CT

Conclusion

Presenting

2:35 - 3:30 CT

Happy Hour

Presenting