2 of 24

Welcome external review committee

Frank Würthwein�Director - SDSC

Saskia de Vries

Principal Product Manager

Allen Institute

Nhan Tran

Wilson Fellow at Fermilab

Patrick Brady

Professor Physics�University of Wisconsin-Milwaukee

V. Ashley Villar

Assistant Professor�Pennsylvania State University

Mitra Taheri

Professor

John Hopkins University

Stephen Neuendorffer

Fellow at AMD Xilinx

3 of 24

Local Host and WIFI

WIFI

Eduroam
SMU Guest
Contact: adeiana@smu.edu

Thanks for our local host!

Allison McCarn Deiana�Assistant Professor�Physics, SMU

4 of 24

https://indico.cern.ch/event/1156222/

5 of 24

Agenda and Venue

Palmer Conference Center, Caruth Hall, 4th floor.

Chamber, Hughes-Triggs Student Center

Bring your own Lunch to poster session!

6 of 24

Community and Safety

Code of Conduct

Be welcoming and open minded
Communicate with empath
Respect each other
Recognize that intent does not equate to impact
Strive to make informal activities relating to A3D3 inclusive to all
Actively work to better our community

7 of 24

Bringing us closer

8 of 24

Our Vision

To establish a tightly coupled organization of domain scientists, computer scientists, and engineers that unite three core components which are essential to achieve real-time AI to transform science: AI techniques, �Computing Hardware, Scientific Applications.

9 of 24

Four Research Areas

Four Research Areas:

Hardware-Algorithm Co-development (HAC)
High Energy Physics (HEP)
Multi-Messenger Astrophysics (MMA)
Neuroscience (Neuros)

10 of 24

Common Computing Challenges

Common in 3 science domains:

HEP: LHC (ATLAS, CMS)
MMA: LIGO, ZTF, DUNE, IceCube
Nero: Neural circuits, Neuro image

Hardware and Algorithm Co-development (HAC) can push the boundaries of data processing beyond industry applications.

As scientific data sets become progressively larger, algorithms to process this data quickly become more complex. In response, AI has emerged as a solution to efficiently analyze these massive data sets. Emerging processor technologies, such as graphics processing units (GPUs) and field-programmable gate arrays (FPGAs), allow AI algorithms to be greatly accelerated. The combination of AI and these processors is leading to a revolution in the way we analyze data, minimizing the time needed to perform the most advanced of analyses, and allowing us to address the challenges brought about by the omnipresent onslaught of data. To harness these new developments and leverage them for the advancement of science, the newly-created $15 million NSF HDR Institute of Accelerated AI Algorithms for Data-Driven Discovery (A3D3) aims to incorporate AI algorithms with new processors such that they can analyze the unprecedented data sets we face.

11 of 24

Collaboration Strategy

Integration of four areas through �two system development:

Heterogeneous
Targeted

Strong synergy to the FastML Lab:

Heterogeneous System

= Coprocessor (Type B)

Targeted System �= Core (Type A)

12 of 24

A Nationwide Institute

9 Institutes, 77 Members

�Members:

17 Senior Personnel
3 Research scientists
11 Postdocs
27 PhD
3 Master
12 Undergrad
4 Postbacs
1 High School

�95% staffed

Program specialist

trainees.html

13 of 24

Equity & Career�J. Duarte

External Advisory Board

M. Bolt (Xilinx)

P. Brady (SCIMMA)�S. Brown (PSC)

M. Cohen (CNBC)

M. Taheri (MCP)

A. Tapper (ICL)

M. Mahoney (Data Science)

P. Mabee (HDR ecosystem)

PI/co-PIs

NeuroAI Integration:�E. Shlizerman

Coordination

HAC : P. Li

HEP : M. Liu �MMA : M. Coughlin (co-PI)

Neuro : A. Orsborn

Management

Director : S.-C. Hsu (PI)

Deputy : P. Harris (co-PI)

Project Manager: to be hired

Executive Board

Approved by NSF, inviting soon

Steering Board

HAC: S. Han

HEP: M. Neubauer (co-PI)

MMA: K. Scholberg (co-PI) Neuro: M. Makin

External: N. Tran, A. Villar� S. Vries

Heterogeneous : B. Riedel

Targeted System : J. Duarte

Engagement : M. Neubauer

Education : M. Graham

Benedikt Riedel

Co-chair TBA

14 of 24

Inter-institute Communication

Institute SLACK channel

FastML Server

Institute Meetings

Agenda:

https://indico.cern.ch/category/14395/

Bi-Weekly

FastML Co-processor (heterogeneous)
FastML HLS4ML (target system)
HAC

Monthly

Neuroscience
Executive Board

Quarterly

Steering board

Annually

Collaboration meeting

15 of 24

External Communication

Website

Starting website:�https://a3d3.ai/
Contract to House of Giants through MIT

Mailing lists

Inquiries: a3d3-engagement@a3d3.ai
Event: a3d3-announcement@a3d3.ai � (83 members)
Internal Institute: a3d3@a3d3.ai
Internal (not frequently used so far):�a3d3-admin, a3d3-hac, a3d3-hep, a3d3-mma, a3d3-neuro, a3d3-heterogeneous, a3d3-targeted-system, a3d3-education, a3d3-equity-career
Social Media

https://www.youtube.com/channel/UCozE47hFawTUb4lUhYlQNIQ (36 subscribes)
https://www.linkedin.com/company/a3d3 (4 followers)
https://twitter.com/a3d3institute (72 follower)

16 of 24

Web Analytics

17 of 24

Kyle

Mia

Pan

Amy

18 of 24

Accomplishments in FY2021

Deliverables

2 papers accepted, 4 papers submitted, 1 talk, 3 posters
6 public sw: TorchSparse v2.0, SPVCNN++, DeepClean, nmma, ScaleHLS, PyLOG

Hiring

1 Institute scientist, 2 Postdocs, 1 Graduate, 4 Postbacs

Outreach & Education

1 event
1 undergraduate course

Institute Events

3 A3D3 seminar: Anqi Wu (Neuros), Zhiru Zhang (HAC), Georgia Karagiorgi (MMA)
2 Executive Board meetings
Annual Collaboration meeting Oct 7, SMU Dallas, TX

Following FastML workshop, Oct 3-6

Community engagement

HDR PI slack channels (invited only) was created
$1.25M supplement grant for HDR Ecosystem activities was submitted

The first HDR PI meeting Oct 26-27, Old Town Alexnadria, VA

19 of 24

Top Priority: FY2022

Review A3D3 management plan

Explicit mandate of each leadership position
Clear strategy of virtual institute collaboration
Student-led cross institutes activity

Reverse site visit in early summer 2022

A3D3 research

We would like your input for the the Year 2 of A3D3

Should we change focus in any way?

21 of 24

Deliverables

Papers/Proceedings

HAC: BEVFusion (submitted to NeurIPS 2022), GSAT (accepted by ICML’22)
HEP: GNN PU removal (2203.15823, submitted to EPJC), MLPF CMS (2203.00330, ACAT 2021)
Targeted system: qonnx (submitted to HiPEAC AccML 2022), TinyML (submitted to MLSys 2022).

Software/Infrastructure

HAC: TorchSparse v2.0, SPVCNN++ (for WOD Challenge 2022)
MMA: DeepClean, nmma
Targeted System: ScaleHLS, PyLog

Applications

CMS Level-1 btagging
RNN jet tagging (paper to be submitted)

Talks

APS CMS ECAL

22 of 24

First Steering Board Review

A3D3 has been online since Oct 1 2021

First quarter

We assembled a strategy document, and hosted a virtual kick-off meeting

Second quarter

We established regular meetings: seminar, research area meetings
We are close to fully staffed: institute research scientists, postbacc

Open position for program operational special

Third quarter

We submitted a $1.25M supplement grant for HDR ecosystem activities
We continue maintained strong research and education activities

We would like your input for the the next quarter

Should we change focus in any way?

23 of 24

News

Website & LOGO

Creating contract with House of Giants via MIT (faster administration process)
LOGO update by Mia

Publication (website)

6 papers
2 talks
1 tutorial

Job opening (website)

Program Operation Specialist: No qualified applicants. Seek for joint hire with other departments
Postdoc (1 position in Duke)
Postbac (4 positions)
Working on hiring Vladimir Loncar (CERN/HLS4ML)

Nice presentation for Feb 14th [indico]

Events

Next A3D3 seminar in 2022

MMA Speaker on Mar 7 (Mon) [Indico]

Next management meeting: Internal Steering Board

The last Friday of each quarter (Mar 25)

24 of 24

Who are we?

HAC

MMA

Neuro

HEP

https://a3d3.ai/researchers.html

Director, PI

co-PI, Dep. Dir.

co-PI

S. Hauck

D. Chen

P. Harris

M. Coughlin

P. Li

S. Han

M. Liu

J. Duart

M. Graham

K.Scholberg

E. Katsavounidis

M. D. Makin

A. Orsborn

K. Hanson

E. Shlizerman

M. Neubauer

S.-C. Hsu