BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Measuring Alignment Between Perceptual Systems: An Analysis Throug
 h The Lens of Shared Invariances - Vedant Nanda\, MPI-SWS + University of 
 Maryland 
DTSTART:20221124T113000Z
DTEND:20221124T123000Z
UID:TALK192845@talks.cam.ac.uk
CONTACT:Adrian Weller
DESCRIPTION:Learning the right invariances is key in learning meaningful r
 epresentations of data. In this talk I will talk about two of our recent w
 orks on measuring if invariances learned by one perceptual model (eg: neur
 al network/human) “align" with another. In the first part of the talk I 
 will talk about alignment of invariances between a neural network and a hu
 man. I will talk about challenges and pitfalls in measuring this alignment
  and will also show some intriguing results about how different choices in
  the deep learning pipeline (architecture\, data augmentation\, loss funct
 ion\, and training paradigm) lead to varying levels of alignment. In the s
 econd part\, I will talk about measuring alignment in invariances between 
 two neural networks. Existing measures that might appear suited for this t
 ask (e.g.\, representation similarity measures) are only narrowly focused 
 on comparing two representations and actually fail to meaningfully capture
  shared invariances between the models that generate these representations
 . I will present our proposal on how to repurpose existing representation 
 similarity methods to faithfully measure shared invariance and will show s
 ome results on how this varies with the choice of network architectures\, 
 loss functions\, random weight initialization\, and datasets used in the t
 raining process. I will close by discussing possible directions for future
  work including using our proposed measure as a constraint during training
 .
LOCATION:Hybrid meeting\, CBL seminar room\, and Zoom https://talks.cam.ac
 .uk/talk/edit/192845
END:VEVENT
END:VCALENDAR