Connectionists: Call for collaboration: Studying gradient norms for batch-size based SGD theory explorations in Horizon Activation Mapping for Time Series Forecasting

[Krupakar Hans]

Fellow researchers,

I have been working on extending gradCAM to time series forecasting over
the past year (https://arxiv.org/pdf/2601.02094) and I would like to
continue to more than with my small GPU. It has been accepted as a
conference paper at the International Conference on Learning and
Optimization (OLA) 2026. Even though the findings have been empirical, I
believe there could be trends in norms by batch sizes that could help with
some claims in gradient descent. If you find the paper interesting or know
someone who would be interested in working with me as leading author(s), I
would be happy to collaborate (10 hours or less / week).

Nevertheless, I believe there is potential in this work and I would like to
have made sure I didn't stop research on this topic. Feel free to work on
it yourselves if you find it interesting and need people who can work more
for the same. My authorship invite would only apply to me and not to my
co-author in the paper because of reasons that don't have to do with the
paper or its potential directions.

Please reach out to me by email for further discussion if you find it
interesting.

Thank you,
Hans K
PhD Candidate,
BITS Pilani Hyderabad Campus,
Telangana, India.



On Mon, 2 Mar, 2026, 5:27 pm Krupakar Hans, <hansk at nyu.edu> wrote:

> Fellow researchers,
>
> I have been working on extending gradCAM to time series forecasting over
> the past year (https://arxiv.org/pdf/2601.02094) and I would like to
> continue to more than with my small GPU. It has been accepted as a
> conference paper at the International Conference on Learning and
> Optimization (OLA) 2026. Even though my findings have been empirical, I
> believe there could be trends in norms by batch sizes that could help with
> some claims in gradient descent. If you find the paper interesting or know
> someone who would be interested in working with me as leading author(s), I
> would be happy to collaborate (10 hours or less / week).
>
> Nevertheless, I believe there is potential in this work and I would like
> to have made sure I didn't stop research on this topic. Feel free to work
> on it yourselves if you find it interesting and need people who can work
> more for the same.
>
> Please reach out to me by email for further discussion.
>
> Thank you,
> Hans K
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20260302/f5236d6c/attachment.html>