NPPS Leadership Team

[Torre Wenaus <wenaus AT gmail.com>]

Hi Brett,

As Elke said your comments read like they are reacting only to the first section, which is served up to us, like a plate of haggis, by CSI. OK that's unfair, I learned in Scotland that I like haggis, especially when it comes with neeps and tatties. 

What do you think of the HEP section, the last section? If you can cast your suggestions of what you want to see into a form suitable for the table in that section, it would be great. Happy Robbie Burns Day.

  Torre

On Mon, Jan 25, 2021 at 3:35 PM Aschenauer Elke-Caroline <elke AT bnl.gov> wrote:

Brett,

I’m must admit lost with your comments are you referring to what Torre added on HEP or are you unhappy about the entire document or something else.

I agree it is not easy to integrate some SCI members in the HEP/NP world to a large extent that come from a different background, and have in 90% of the cases a very different language.

My understanding was this exercise is to describe what is needed to change this.

If I look in NP there are limited AI/ML experts who are working in physics, so indeed for EIC we can use more manpower working with us co-designing AI/ML in our detectors. Maybe we can find all this manpower in HEP at physics, perfect and I’m open to collaborate with everybody who is interested in EIC.

Cheers elke

On Jan 25, 2021, at 15:18, Brett Viren <bv AT bnl.gov> wrote:

Hi Torre,

It is really hard to interpret this vision into my frame of reality but
what I do interpret is very negative.  Hopefully I'm not too blunt but
here is what I read:

- Shields up around CSI/SDCC in form of some abstract interface
 separating facility resources from experiment users.

- Interface to be defined by CSI/SDCC and experiments must figure if/how
 it can be used.

- Current BNLers who are AI/ML-adjacent suck at AI/ML, hire new ones and
 put them in CSI.

- Build some vauge automation to solve unspecified problems, but mostly
 facilities related ones.

I hope my interpretation is just plain wrong as this is a pretty bleak
picture.

What I want to see is something that hits these points:

- Deeper integration of CSI expertise and personnel into BNL
 experiments.  CSI meets the domains, not the domains bow to the gates
 of CSI.  CSI scientists work alongside domain scientists to explore
 new ways to apply AI/ML to solve domain problems.  (We are having some
 amount of success now with this approach!)

- Tailored interfaces between SDCC and BNL experiments.  While providing
 those, SDCC seeks and applies ways for reuse and automation and
 resources sharing across its customers.

- Lowered/zeroed bar to BNL domain scientists use of GPU and other AI/ML
 hardware to enable uplifting of skills and development of new AI/ML
 techniques and their application.  Ie, like RACF but with GPU/HPC.
 Sure, require some "proposal" for large production running but if I
 have to defend a proposal for a few GPUs, I'll buy them myself - or
 more likely drop my idea and work on the myriad of other things going
 on.

- Recognize that most of the effort in applying AI/ML is domain-specific
 data wrangling work.  Once in a form AI/ML codes can accept, the hard
 work is done and the play time begins.  Given the variety of needs,
 putting up interfaces to infrastructure will only exacerbate what
 domain scientists need to do in the wrangling.  Better to have CSI
 work with domain scientists to educate them in ways to do the data
 wrangling, where needed, or have CSI actually contribute effort
 directly to that wrangling.  In doing so, common interfaces that
 actually work the way the scientists do may be developed to improve
 future data wrangling.

- Tie this 5-year "autonomous" thing to some thing more useful and
 concrete.  Simply automating facilities so facilities can facilitate
 is a poor vision.  Maybe without naming names ATLAS's automated
 production error response system or DUNE self-healing DAQ ideas can be
 invoked.


-Brett.


Torre Wenaus <wenaus AT gmail.com> writes:

About the lab-wide efforts/goals in AI/ML... Frank Alexander is
renewing the effort to put something together and I was asked to
add a HEP part to this googledoc (it's a copy, Frank restricted
access to the original). My section follows one from Elke, and
Jerome has been asked for input also. We were each asked to produce
a version of the table at the top of the document. Comments
appreciated.

https://docs.google.com/document/d/1au5ywJU1gvfLhqWr-n1ol0dxUYtBxdZchwMb-oWxui4/edit?usp=sharing

  Torre

On Thu, Jan 7, 2021 at 1:41 PM Torre Wenaus <wenaus AT gmail.com>
wrote:

   Darn.

   Thanks, Brett! Added

   On Thu, Jan 7, 2021 at 1:36 PM Michael Begel <mbegel AT gmail.com>
   wrote:

       Nope!

       Michael Begel

       On Thu, Jan 7, 2021 at 1:32 PM Torre Wenaus <
       wenaus AT gmail.com> wrote:

           Beats me! Do you know?

           On Thu, Jan 7, 2021 at 1:29 PM Michael Begel <
           mbegel AT gmail.com> wrote:

               One comment is that I don't see the connection
               between the work outlined and the *lab's* goals in
               AI/ML. What are the lab-wide efforts and goals?

               Michael Begel

               On Thu, Jan 7, 2021 at 1:21 PM Torre Wenaus <
               wenaus AT gmail.com> wrote:

                   Draft slides are here, comments appreciated...

                   https://docs.google.com/presentation/d/1f5g4IXFS2_crHVfwf-rsEMseHc31V_2cOxJrs_0t884/edit?usp=sharing

                     Torre

                   On Tue, Jan 5, 2021 at 12:26 PM Brett Viren <
                   bv AT bnl.gov> wrote:

                       Torre Wenaus via Phys-npps-mgmt-l <
                       phys-npps-mgmt-l AT lists.bnl.gov>
                       writes:

And I have Brett's LDRD.

                       In addition to the "LS4GAN" LDRD which
                       pushes on a "new" (really a
                       variant on existing) ML technique, some
                       established techniques have been
                       applied by Haiwang Yu (EDG postdoc) on two
                       problems:

                       1) Improve the ionization signal
                       reconstruction efficiency in LArTPC for
                       particularly problematic track topology. 
                       (Paper accepted to JINST).

                       2) Neutrino vertex finding in a sea of
                       cosmic muons in the LArTPC
                       MicroBooNE detector on the surface.

                       -Brett.

                   --
                   -- Torre Wenaus, BNL NPPS Group, ATLAS
                   Experiment
                   -- BNL 510A 1-222 | 631-681-7892 |  
                   wenaus AT gmail.com | npps.bnl.gov | wenaus.com
                   -- NPPS Mattermost room:
                   https://chat.sdcc.bnl.gov/npps/channels/town-square

           --
           -- Torre Wenaus, BNL NPPS Group, ATLAS Experiment
           -- BNL 510A 1-222 | 631-681-7892 |  wenaus AT gmail.com | 
           npps.bnl.gov | wenaus.com
           -- NPPS Mattermost room:
           https://chat.sdcc.bnl.gov/npps/channels/town-square

   --
   -- Torre Wenaus, BNL NPPS Group, ATLAS Experiment
   -- BNL 510A 1-222 | 631-681-7892 |  wenaus AT gmail.com | 
   npps.bnl.gov | wenaus.com
   -- NPPS Mattermost room:
   https://chat.sdcc.bnl.gov/npps/channels/town-square

--
-- Torre Wenaus, BNL NPPS Group, ATLAS Experiment
-- BNL 510A 1-222 | 631-681-7892 |  wenaus AT gmail.com | npps.bnl.gov
 | wenaus.com
-- NPPS Mattermost room:
https://chat.sdcc.bnl.gov/npps/channels/town-square

Note: Please do not feel obligated to respond to this message outside of your work hours.

 ( `,_' )+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-

  )    `\                                                      +

 /    '. |        Elke-Caroline Aschenauer  (she/her/hers)      =

 |       `,                                                      -

  \,_  `-/        Brookhaven National Lab                         +

  ,&&&&&V         Physics Dept.,            25 Corona Road         =

 ,&&&&&&&&:       Bldg. 510 /2-195          Rocky Point, NY,        -

,&&&&&&&&&&;      20 Pennsylvania Avenue                 11778       +

|  |&&&&&&&;\     Upton, NY 11973                                     =

|  |       :_) _  Tel.:  001-631-344-4769   Tel.:   +1-631-569-4290    -

|  |       ;--' |                           Cell-USA:  +1-757-2565224   +

'--'   `-.--.   |                           Cell-Europe: +49152220025211 =

   \_    |  |—‘                                                           -

     `-._\__/     Mail: elke AT bnl.gov        elke.caroline AT me.com           +

            +=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=-+=

--

-- Torre Wenaus, BNL NPPS Group, ATLAS Experiment

-- BNL 510A 1-222 | 631-681-7892 |  wenaus AT gmail.com | npps.bnl.gov | wenaus.com

-- NPPS Mattermost room: https://chat.sdcc.bnl.gov/npps/channels/town-square