NPPS Leadership Team

[Brett Viren <bv AT bnl.gov>]

Thanks Torre,

I read through the summary and description of the RAs.  If we attempt to
exploit/apply what we learned from the LS4GAN LDRD I think RS3 is the
best fit.  RA1 is a maybe.  RA2 is a stretch.  The last two I do not see
a connection.

I typed some rambling notes as I read.  They are below.

-Brett.


RA1: I feel this research area is hoping to invent some new class of DL
akin to LLM.  LS4GAN's network (UVCGAN) currently does not come close to
being a "foundational model" but maybe LS4GAN techniques could touch on
some of the (sub) goals of RA1:

- Growing a much larger LS4GAN-type network than current UVCGAN and
  tasking it to translate between many, vastly different domains may get
  into "foundational model" space.

- Working to solve the "LS4GAN Dilemma" might fall into "significantly
  advance the state of the art in computational science".  In
  particular, I think we must find a way to construct a metric space in
  the otherwise abstract latent space of the network if we are to apply
  the general domain translation concept with scientific rigor needed to
  estimate systematic uncertainties.

- The basic (and working, if subject to the dilemma) function of the
  LS4GAN technique is right on target to "minimize risk, model misuse,
  unintentional bias or inaccuracies".  Addressing these were what
  motivated the concept to start with.

- Training UVCGAN for initial "simple" toy LArTPC problems we concocted
  used dozens of GPUs for weeks for each training push.  We have a small
  list of effects that UVCGAN failed to learn which may be due to lack
  of data and thus may be captured with more CPU/GPU.  As we might move
  from "toy" to more real problems a story that involves substantial HPC
  resources might be concocted.

RA2: Maybe there is a case to say LS4GAN translations, especially with a
metric latent space or otherwise solving the LS4GAN Dilemma, is a form
of "knowledge synthesis".  LS4GAN certainly addresses the criticism that
current techniques "do not produce ... verified, uncertainty-quantified
... results.".

RA3: LS4GAN ideas fit squarely into the "outer loop" discussion.  Couple
LS4GAN with a differentiable physics/detector simulation and it hits
even more on target.


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

> Hong has already asked Alexei and me about ideas we can pursue... this one 
> is worth looking at and thinking about
>
> ---------- Forwarded message ---------
> From: Kleese Van Dam, Kerstin <kleese AT bnl.gov>
> Date: Tue, Feb 13, 2024 at 1:16 PM
> Subject: Re: ASCR call for proposals in AI for Science
> To: CSI Directorate <CSIDirectorate AT bnl.gov>, Misewich, James 
> <misewich AT bnl.gov>, McSweeney, Elspeth <emcsweeney AT bnl.gov>,
> Wilkins, Stuart <swilkins AT bnl.gov>, Mason, Sara <smason AT bnl.gov>, Gao, 
> Haiyan <hgao AT bnl.gov>, Torre Wenaus <
> wenaus AT gmail.com>, Klimentov, Alexei <aak AT bnl.gov>, Schoonen, Martin 
> <mschoonen AT bnl.gov>, Yager, Kevin <kyager AT bnl.gov>,
> Cutler, Cathy <ccutler AT bnl.gov>, Yeck, Jim <jyeck AT bnl.gov>, Aschenauer, 
> Elke <elke AT bnl.gov>, Asner, David <dasner AT bnl.gov>
> Cc: Hill, John <hill AT bnl.gov>, Kleese Van Dam, Kerstin <kleese AT bnl.gov>
>
> Dear all,
>
> A correction the submission limit is 3 pre-proposals per topic so 15 in 
> total for BNL, we will still coordinate the
> responses and all other details were correct.
>
> Best wishes,
>
> Kerstin
>
>  
>
> Kerstin Kleese Van Dam
> Director
> Computational Science Initiative
>
> Building 725, room 2-127B
> Office: 631.344.6019
>
> Cell: 509 221 0758
> kleese AT bnl.gov
>                        
> signature_4173528834   
> Twitter | Facebook | Instagram | LinkedIn
>
>  
>
>  
>
>  
>
> From: Kleese Van Dam, Kerstin <kleese AT bnl.gov>
> Date: Tuesday, February 13, 2024 at 9:57 AM
> To: CSI Directorate <CSIDirectorate AT bnl.gov>, Misewich, James 
> <misewich AT bnl.gov>, McSweeney, Elspeth <emcsweeney AT bnl.gov>,
> Wilkins, Stuart <swilkins AT bnl.gov>, Mason, Sara <smason AT bnl.gov>, Gao, 
> Haiyan <hgao AT bnl.gov>, Torre Wenaus <
> wenaus AT gmail.com>, Klimentov, Alexei <aak AT bnl.gov>, Schoonen, Martin 
> <mschoonen AT bnl.gov>, Yager, Kevin <kyager AT bnl.gov>,
> Cutler, Cathy <ccutler AT bnl.gov>, Yeck, Jim <jyeck AT bnl.gov>, Aschenauer, 
> Elke <elke AT bnl.gov>, Asner, David <dasner AT bnl.gov>
> Cc: Hill, John <hill AT bnl.gov>, Kleese Van Dam, Kerstin <kleese AT bnl.gov>
> Subject: ASCR call for proposals in AI for Science
>
> Dear all,
>
> Today ASCR released its AI for Science call for this year. This call is 
> limited to 3 applications led per lab, for that
> reason all submissions will be coordinated through CSI. If you are 
> interested in submitting, please be so kind to submit a
> pre-proposal to CSIProposals AT bnl.gov using the attached form. Please feel 
> free to distribute this message, we are
> welcoming submissions from all BNL directorates.
>
> Advancements in Artificial Intelligence for Science
>
> Close Dates:
>
>   ● BNL internal pre-application COB March 1, 2024
>   ● Go-noGo decisions provided March 8, 2024
>   ● Pre-applications are due March 19, 2024 at 5:00 PM Eastern Time.
>   ● Pre-application Response date is April 11, 2024 at 11:59 PM Eastern 
> Time.
>
> The DOE SC program in Advanced Scientific Computing Research (ASCR) hereby 
> announces its interest in basic computer
> science and applied mathematics research in the fundamentals of Artificial 
> Intelligence (AI) for science. Specifically,
> advancements in this area are sought that can enable the development of:
>
>   ● Foundation models for computational science;
>   ● Automated scientific workflows and laboratories;
>   ● Scientific programming and scientific-knowledge-management systems;
>   ● Federated and privacy-preserving training for foundation and other AI 
> models for science; and
>   ● Energy-efficient AI algorithms and hardware for science.
>
> The development of new AI techniques applicable to multiple scientific 
> domains can accelerate progress, increase
> transparency, and open new areas of exploration across the scientific 
> enterprise.
>
>   ● Budget: for National Labs $250K-$2M/year all other applicants 
> $100K-$350K/year
>   ● Project duration 3 Years
>   ● Overall funding available is $36M for 3 years, so $12M/year available 
> for projects.
>
>  Best wishes,
>
> Kerstin
>
>  
>
>  
>
> Kerstin Kleese Van Dam
> Director
> Computational Science Initiative
>
> Building 725, room 2-127B
> Office: 631.344.6019
>
> Cell: 509 221 0758
> kleese AT bnl.gov
>                        
> signature_2101736385   
> Twitter | Facebook | Instagram | LinkedIn
>
>  
>
>  
>
> --
> -- Torre Wenaus, BNL NPPS Group, ATLAS Experiment
> -- BNL 510A 1-222 | 631-681-7892
> _______________________________________________
> Phys-npps-mgmt-l mailing list
> Phys-npps-mgmt-l AT lists.bnl.gov
> https://lists.bnl.gov/mailman/listinfo/phys-npps-mgmt-l

Attachment: signature.asc
Description: PGP signature