From awd at cs.cmu.edu  Fri Apr  3 06:47:25 2026
From: awd at cs.cmu.edu (Artur Dubrawski)
Date: Fri, 3 Apr 2026 06:47:25 -0400
Subject: If you do multimodal or vision AI you should check this out
Message-ID: <CAJvAoyuMk7qU=u1xf5X-eerqe6VuH0+XLftie2-OPMn47KFROg@mail.gmail.com>

https://x.com/heygurisingh/status/2039012548260082082?s=20
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/autonlab-users/attachments/20260403/f88d1796/attachment.html>

From awd at cs.cmu.edu  Fri Apr  3 18:34:08 2026
From: awd at cs.cmu.edu (Artur Dubrawski)
Date: Fri, 3 Apr 2026 18:34:08 -0400
Subject: Fwd: If you do multimodal or vision AI you should check this out
In-Reply-To: <CAPYU4y+8dMriYUaHdpog2fJhH___gEJe9szGDtv15zW0ksFGTQ@mail.gmail.com>
References: <CAJvAoyuMk7qU=u1xf5X-eerqe6VuH0+XLftie2-OPMn47KFROg@mail.gmail.com>
 <CAPYU4y+OOwmR56dzGQBc1zLbjk87312rB__6C7h88ywwWM=8Gg@mail.gmail.com>
 <CAPYU4y+8dMriYUaHdpog2fJhH___gEJe9szGDtv15zW0ksFGTQ@mail.gmail.com>
Message-ID: <CAJvAoyuU+V+NB6uBNj9hf9arqcwObFX5aMemPbgpQJRU1337Pg@mail.gmail.com>

Sharing Szymon's insights as they have a broader appeal I think.

Also, the paper confirms our much earlier observation that AI benchmarks do
not measure up to the models they are supposed to assess.
That's why we have invested time and effort to develop benchmarking
*frameworks* that would let us dynamically generate new benchmarks
that would hopefully be able to stay ahead of the capabilities of the AI
technology as it continues to evolve.
Basically, putting the horses in front of the carriage again. Big thanks to
TimeSeriesGym and TimeSeriesExamAgent teams
for spearheading these efforts here at the Auton Lab!

Cheers,
Artur

PS It is hard to blame an AI model for accomplishing their tasks with
whatever we give them.
It was always the case in ML that we should be careful about how we (or our
AI agents these days) test the models properly,
to make sure they are doing their things in the ways we expect them to do.

---------- Forwarded message ---------
From: Szymon Rusiecki <srusieck at andrew.cmu.edu>
Date: Fri, Apr 3, 2026 at 9:57?AM
Subject: Re: If you do multimodal or vision AI you should check this out
To: Artur Dubrawski <awd at cs.cmu.edu>


After reproducing methodology presented in their paper, the mirage issue
occurs only for ?big?
models. The ?small? ones often don?t have this issue.

On Fri, Apr 3, 2026 at 15:41 Szymon Rusiecki <srusieck at andrew.cmu.edu>
wrote:

> I am actually surprised as I recently broke my collarbone so I decided to
> test on Gemini 3 flash with OOD sample (I think Google doesn?t have an
> image from my iPhone and even if, the photo doesn?t have any description)
> with prompt ?what do you see on this image?? and it responded with the same
> answer as my doctor.
>
> SR
>
> On Fri, Apr 3, 2026 at 12:48 Artur Dubrawski <awd at cs.cmu.edu> wrote:
>
>> https://x.com/heygurisingh/status/2039012548260082082?s=20
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/autonlab-users/attachments/20260403/a7a48f77/attachment.html>

From awd at cs.cmu.edu  Tue Apr  7 14:48:36 2026
From: awd at cs.cmu.edu (Artur Dubrawski)
Date: Tue, 7 Apr 2026 14:48:36 -0400
Subject: Fwd: RI PhD Thesis Proposal - Xinyu (Rachel) Li
In-Reply-To: <CABKA2NUikVDpsQY7dGrXxgaNRFLTROCpeUy89De2rOwfkJjzFA@mail.gmail.com>
References: <CABKA2NUikVDpsQY7dGrXxgaNRFLTROCpeUy89De2rOwfkJjzFA@mail.gmail.com>
Message-ID: <CAJvAoyuExUKJt5=LX863juW22SLMeqefwobuLsVKUHnSj=AtbQ@mail.gmail.com>

Team,

Please come and witness Rachel giving her excellent thesis proposal
presentation on Monday next week!

Cheers,
Artur

---------- Forwarded message ---------
From: RI PhD Program Manager <ri-phd-manager at andrew.cmu.edu>
Date: Tue, Apr 7, 2026 at 2:45?PM
Subject: RI PhD Thesis Proposal - Xinyu (Rachel) Li
To: RI People <ri-people at andrew.cmu.edu>


*RI CALENDAR EVENT
<https://www.ri.cmu.edu/event/ri-phd-thesis-proposal-xinyu-rachel-li/>*




*Date: April 13, 2026Time: 03:15 PM (ET) Location: GHC 6121Zoom Link
<https://cmu.zoom.us/j/96249846505?pwd=aZalDPGdL4JsUbJR0M0YQOPta8pFZX.1>*

*Type: Ph.D. Thesis Proposal*
*
<https://cmu.zoom.us/j/96249846505?pwd=aZalDPGdL4JsUbJR0M0YQOPta8pFZX.1>Who:
Xinyu (Rachel) Li*
*Title: Towards Accessible AI Agents*

*Abstract:*
Empowered by large language models (LLMs), AI agents have shown strong
potential across tasks such as general-purpose assistance, software coding,
and scientific research. However, their practical utility in applications
involving consequential decisions such as healthcare, remains constrained
by three major challenges.

*Evaluation.* Existing agent evaluations often focus on well-structured
tasks and final outcomes, failing to fully capture the complexity of
real-world workflows. We propose evaluation frameworks grounded in
realistic machine learning engineering workflows, providing skill-based,
multi-artifact, and holistic assessments that systematically evaluate the
practical utility of AI agents.

*Learning.* Improving LLMs for agentic use typically relies on
reinforcement learning with large amounts of high-quality labeled data,
which are costly and difficult to obtain in expert domains including
healthcare. To address this limitation, we aim to develop learning
frameworks that require minimal external supervision, improving the
scalability and efficiency of agent learning.

*Specialization.* AI agents typically follow a one-size-fits-all paradigm
at the time of deployment, lacking mechanisms to account for task-specific
or user-specific requirements. We propose methods that enable agent
specialization for downstream tasks and users, expanding their
applicability across heterogeneous deployment settings.

This thesis aims to make AI agents more broadly accessible and impactful in
important real-world applications by enhancing their practical utility,
making them more measurable, more capable, and better tailored to the needs
of their users and applications.

*Link to thesis*
<https://drive.google.com/drive/folders/1mPlZXQ3WLa42e1LFKyMkjuHy_diDc6yy?usp=sharing>

*Thesis committee members:*
Artur Dubrawski (Chair)
Andrea Bajcsy
Barnab?s P?czos
Daniel McDuff (Google)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/autonlab-users/attachments/20260407/9cf189e7/attachment.html>

From awd at cs.cmu.edu  Mon Apr 27 14:09:19 2026
From: awd at cs.cmu.edu (Artur Dubrawski)
Date: Mon, 27 Apr 2026 14:09:19 -0400
Subject: Mark your calendars! [RI PhD Thesis Proposal - Willa Potosnak]
Message-ID: <CAJvAoyuhGfdyjGRsKAeLH8Z9xH9dDu=Pm1Hv_OhF22TXhqpUgQ@mail.gmail.com>

Willa will be giving her proposal presentation on Monday next week during
our usual brainstorming session time slot, but in Gates Hall. Please come
and join. Intellectual fun guaranteed!

Cheers,
Artur

---------- Forwarded message ---------
From: RI PhD Program Manager <ri-phd-manager at andrew.cmu.edu>
Date: Mon, Apr 27, 2026, 1:52?PM
Subject: RI PhD Thesis Proposal - Willa Potosnak
To: RI People <ri-people at andrew.cmu.edu>


*RI EVENT CALENDAR*
<https://www.ri.cmu.edu/event/forecasting-at-scale-with-efficient-deep-learning-architectures/>

*Date:  *May 4th, 2026
*Time:  *4:30 PM-6:00 PM
*Location*: GHC 4405
*Zoom Link*
<https://cmu.zoom.us/j/93920625206?pwd=WsTYphnRXRfhUaBpKKbbeIaTgDMoro.1>
*Type: *RI PhD Thesis Proposal
*Who: *Willa Potosnak

*Title:* Forecasting at Scale with Efficient Deep Learning Architectures

*Abstract:*
Time Series Foundation Models (TSFMs) have scaled rapidly, with publicly
reported pretraining corpora growing from 1.23 billion to 1 trillion data
points between 2024 and 2026, an approximately 800? increase in two years.
Recent work has further supplemented real-world data with synthetic data to
expose models to broader time series patterns. Yet, this data-centric
paradigm raises a fundamental question: *must intelligent forecasting rely
solely on scale, or can intentional architectural design unlock better
generalization? *This thesis proposes that more intelligently and
efficiently leveraging existing data, rather than scale alone, is key to
achieving better forecasting generalization. We pursue this through three
parallel architectural themes: exploiting cross-channel structure beyond
temporal patterns, enabling zero-shot generalization through structured
composition, and reducing gradient and forecast variance by design. Each
theme aims to enhance generalization with available data while treating
computational efficiency as a core design principle.

In this thesis, we demonstrate that scale is not the only path to
generalization by: developing multivariate architectures that leverage
cross-channel dependencies efficiently while reducing forecast error;
showing that architectures can generalize beyond their training
distribution in both patterns and concepts; and verifying variance-aware
architectural designs that extract richer training signals from existing
data, provably reducing gradient variance while reducing forecast error and
improving calibration.

Within the first theme, we further propose pretraining strategies for
multivariate TSFMs to investigate whether data balancing and curriculum
learning can improve downstream generalization given the same pretraining
corpora. Within the second theme, we propose an additional dimension of
generalization, extending beyond pattern and concept generalization to
horizon generalization, an important consideration for TSFMs applied across
diverse tasks and domains. Overall, this work contributes new insights into
advancing time series forecasting generalization through efficient
architectural design.

*Committee:*
Artur Dubrawski, Chair
John Dolan
Barnab?s P?czos
Michael W. Mahoney (University of California, Berkeley)

*Thesis Link*
<https://drive.google.com/file/d/1oYYdgxvLaW4iF-LWHqzpWHMuXA677QYv/view?usp=sharing>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/autonlab-users/attachments/20260427/383e0c0c/attachment.html>