Meenakshi Khosla
@meenakshik93
Followers
918
Following
399
Statuses
93
Assistant Professor @UCSD @CogSci | Past: Postdoctoral researcher @MIT , Phd @Cornell, BTech @IITKanpur | Interested in biological and artificial intelligence
Cambridge, MA
Joined January 2018
Super excited to share our recent work on privileged representational axes in biological and artificial neural networks! w/ @Nancy_Kanwisher
@JoshHMcDermott @ItsNeuronal
6
58
191
@mtoneva1 Very interesting! I was also reminded of your older paper showing a similar positive impact of explicit alignment with brain data in the language domain
1
0
0
RT @eghbal_hosseini: Why do diverse ANNs resemble brain representations? Check out our new paper with @_coltoncasto, @NogaZaslavsky, Colin…
0
49
0
RT @unnatjain2010: Excited to share that I'll be joining University of California at Irvine as a CS faculty in '25!🌟 Faculty apps: @_krish…
0
71
0
RT @ItsNeuronal: Very happy to share a new paper that will appear in UniReps proceedings (. I show how two popular…
0
44
0
RT @nacloos: ⁉️What do model-neural similarity scores tell us? To systematically explore this for different metrics, we develop new numeri…
0
58
0
RT @Ansh_soni1234: @aran_nayebi @jeffrey_bowers @RylanSchaeffer Hi @aran_nayebi, just wanted to clarify that the paper @jeffrey_bowers has…
0
3
0
To your last point, strongly agree. The benefits of predictive modeling and linear predictivity as a metric are well known in the field, and extend beyond model-brain comparisons (eg. neural population control, in-silico experiments on large datasets to probe response properties). I myself use linear predictivity all the time! The limitations however are less acknowledged explicitly and discussed, so i think its great we are having these discussions :)
0
0
1
e.g. in comparing different architectural families (CNNs vs transformers) with various metrics shows that only soft-matching distinguishes them, revealing similarities in population representations and differences in neural tuning
0
0
5
Representational comparisons are often used to draw conclusions about the functional correspondence between systems. In this context, choosing similarity measures that better align with functional behaviors (e.g. CKA, Procrustes) can be more informative (see paper linked above)
1
0
3
RT @PaulunVivian: I’m excited to share that I will join @UWPsych as an Assistant Professor starting in Fall 2025. 🎇🥳 I feel incredibly fort…
0
23
0
RT @JoshHMcDermott: Please RT: I am looking to hire a research assistant to help recruit, schedule and run human participants in auditory p…
0
121
0
Check out Ansh's impressive work demonstrating how the choice of metrics influences the conclusions drawn from model-brain comparisons. Keep an eye on Ansh — he’s set to make significant contributions in the NeuroAI space!
To ask how similar the brain is to a neural network we need a similarity metric. In a new paper I asked how much the metric matters to downstream conclusions, and, upshot, it matters a great deal. (1/7)
1
5
30
@lowrank_adrian @Nancy_Kanwisher @JoshHMcDermott @ItsNeuronal great question! one starting point could be to use this framework to isolate tuning functions that are significantly more prominent in the native axes compared to random axes, and focus on those
0
0
1