Venkat Ramaswamy
@venkramaswamy.bsky.social
Theoretical Neuroscience, Deep Learning, & the space between.
Assistant Professor, Birla Institute of Technology & Science.
http://brain.bits-hyderabad.ac.in/venkat/
Assistant Professor, Birla Institute of Technology & Science.
http://brain.bits-hyderabad.ac.in/venkat/
We asked if we could similarly leverage the subspace geometry to obtain robustness to adversarial attacks. We built variants of MASC that ended up being upto ~3x better than the model, even though both use the same initial substrate.
More here: openreview.net/pdf?id=0MWW5...
(10/N)
More here: openreview.net/pdf?id=0MWW5...
(10/N)
October 24, 2025 at 5:25 PM
We asked if we could similarly leverage the subspace geometry to obtain robustness to adversarial attacks. We built variants of MASC that ended up being upto ~3x better than the model, even though both use the same initial substrate.
More here: openreview.net/pdf?id=0MWW5...
(10/N)
More here: openreview.net/pdf?id=0MWW5...
(10/N)
So, what’s an adversarial attack?
It turns out that with typical Deep Networks a malicious adversary can change inputs, e.g. images so the Deep Net classifies it as something else. See the example image below from blog.mi.hdm-stuttgart.de/index.php/20...
(8/N)
It turns out that with typical Deep Networks a malicious adversary can change inputs, e.g. images so the Deep Net classifies it as something else. See the example image below from blog.mi.hdm-stuttgart.de/index.php/20...
(8/N)
October 24, 2025 at 5:12 PM
So, what’s an adversarial attack?
It turns out that with typical Deep Networks a malicious adversary can change inputs, e.g. images so the Deep Net classifies it as something else. See the example image below from blog.mi.hdm-stuttgart.de/index.php/20...
(8/N)
It turns out that with typical Deep Networks a malicious adversary can change inputs, e.g. images so the Deep Net classifies it as something else. See the example image below from blog.mi.hdm-stuttgart.de/index.php/20...
(8/N)
Paper #2 w/
@simran-ketha.bsky.social
@mummani-nuthan.bsky.social
&
@niranjanrajesh.bsky.social
Here we considered the setting of adversarial attacks.
(7/N)
@simran-ketha.bsky.social
@mummani-nuthan.bsky.social
&
@niranjanrajesh.bsky.social
Here we considered the setting of adversarial attacks.
(7/N)
October 24, 2025 at 5:12 PM
Paper #2 w/
@simran-ketha.bsky.social
@mummani-nuthan.bsky.social
&
@niranjanrajesh.bsky.social
Here we considered the setting of adversarial attacks.
(7/N)
@simran-ketha.bsky.social
@mummani-nuthan.bsky.social
&
@niranjanrajesh.bsky.social
Here we considered the setting of adversarial attacks.
(7/N)
Surprisingly, we find that this works extraordinarily well.
For every model tested, on at least one layer MASC beats the model test accuracy, and in many cases by a significant margin (see table below), and especially for cases where there is high degree of corruption of labels.
(5/N)
For every model tested, on at least one layer MASC beats the model test accuracy, and in many cases by a significant margin (see table below), and especially for cases where there is high degree of corruption of labels.
(5/N)
October 24, 2025 at 5:12 PM
Surprisingly, we find that this works extraordinarily well.
For every model tested, on at least one layer MASC beats the model test accuracy, and in many cases by a significant margin (see table below), and especially for cases where there is high degree of corruption of labels.
(5/N)
For every model tested, on at least one layer MASC beats the model test accuracy, and in many cases by a significant margin (see table below), and especially for cases where there is high degree of corruption of labels.
(5/N)
Excited to have 2 papers accepted at @neuripsconf.bsky.social 2025 - Reliable ML Workshop.
Paper #1: openreview.net/forum?id=9Ue...
Paper #2: openreview.net/forum?id=0MW...
I’ll briefly summarize the work below.
(Links above if you want to get right to the papers)
THREAD 🧵
Please RT.
(1/N)
Paper #1: openreview.net/forum?id=9Ue...
Paper #2: openreview.net/forum?id=0MW...
I’ll briefly summarize the work below.
(Links above if you want to get right to the papers)
THREAD 🧵
Please RT.
(1/N)
October 24, 2025 at 5:08 PM
Excited to have 2 papers accepted at @neuripsconf.bsky.social 2025 - Reliable ML Workshop.
Paper #1: openreview.net/forum?id=9Ue...
Paper #2: openreview.net/forum?id=0MW...
I’ll briefly summarize the work below.
(Links above if you want to get right to the papers)
THREAD 🧵
Please RT.
(1/N)
Paper #1: openreview.net/forum?id=9Ue...
Paper #2: openreview.net/forum?id=0MW...
I’ll briefly summarize the work below.
(Links above if you want to get right to the papers)
THREAD 🧵
Please RT.
(1/N)
September 1, 2025 at 11:31 AM
"Early to bed, early to rise, work like hell, and advertise."
-Ted Turner
-Ted Turner
August 23, 2025 at 5:34 PM
"Early to bed, early to rise, work like hell, and advertise."
-Ted Turner
-Ted Turner
Today, I modified the definition of plagiarism to keep with the times
February 15, 2025 at 3:05 PM
Today, I modified the definition of plagiarism to keep with the times
If you've been waiting for a sign, this is it
January 26, 2025 at 2:30 PM
If you've been waiting for a sign, this is it
Interesting point here. How do we better visualize/hear/feel the data in real-time, so we can quickly pick follow-up experiments?
December 15, 2024 at 4:22 PM
Interesting point here. How do we better visualize/hear/feel the data in real-time, so we can quickly pick follow-up experiments?
This will be Poster #68 at the NeurIPS Attrib Workshop today in Meeting Room 205-207, Vancouver Convention Center at 3:00pm-4:30pm PST.
Not able to make it due to visa issues, but we will be live just a QR code away.
(w/ @mummani-nuthan.bsky.social @simran-ketha.bsky.social)
Not able to make it due to visa issues, but we will be live just a QR code away.
(w/ @mummani-nuthan.bsky.social @simran-ketha.bsky.social)
December 14, 2024 at 5:10 PM
This will be Poster #68 at the NeurIPS Attrib Workshop today in Meeting Room 205-207, Vancouver Convention Center at 3:00pm-4:30pm PST.
Not able to make it due to visa issues, but we will be live just a QR code away.
(w/ @mummani-nuthan.bsky.social @simran-ketha.bsky.social)
Not able to make it due to visa issues, but we will be live just a QR code away.
(w/ @mummani-nuthan.bsky.social @simran-ketha.bsky.social)
If you've been looking for a sign, this is it
December 12, 2024 at 12:57 PM
If you've been looking for a sign, this is it
Nice. I have a similar one above my desk too!
November 28, 2024 at 4:26 AM
Nice. I have a similar one above my desk too!
Blue bird is good
November 16, 2024 at 4:32 AM
Blue bird is good
Having fun dabbling with a low-profile mechanical keyboard
November 15, 2024 at 5:15 PM
Having fun dabbling with a low-profile mechanical keyboard
I thought Sydney Brenner himself was the source of this dogma; see excerpt below from
www.cell.com/current-biol...
www.cell.com/current-biol...
November 12, 2024 at 5:00 PM
I thought Sydney Brenner himself was the source of this dogma; see excerpt below from
www.cell.com/current-biol...
www.cell.com/current-biol...
Apropos of nothing
November 10, 2024 at 3:36 AM
Apropos of nothing
Atypical primate visitor at my office
November 9, 2024 at 4:24 PM
Atypical primate visitor at my office
No guts, no glory
October 30, 2024 at 6:14 PM
No guts, no glory
I'm feeling lucky
October 30, 2024 at 11:34 AM
I'm feeling lucky
This was part of the Masters' thesis of the extraordinary
@mummani-nuthan.bsky.social
who is applying for Ph.D. positions next Fall.
(10/n)
@mummani-nuthan.bsky.social
who is applying for Ph.D. positions next Fall.
(10/n)
October 28, 2024 at 2:42 PM
This was part of the Masters' thesis of the extraordinary
@mummani-nuthan.bsky.social
who is applying for Ph.D. positions next Fall.
(10/n)
@mummani-nuthan.bsky.social
who is applying for Ph.D. positions next Fall.
(10/n)