New paper: turns out you can train deep nets without normalization layers by replacing them with a parameterized tanh()
@ylecun
-
Nvidia Chief Scientist Bill Dally Fireside Chat at GTC
By
–
I'll be doing a fireside chat at GTC with Nvidia chief scientist Bill Dally Tuesday next week.
-
NSF NIH Grant Funding Challenges for AI Research
By
–
(a) [citation needed]
(b) Do you know how ridiculously competitive NSF and NIH grant applications are? Only a small proportion of applications get funded, less than 20%. -
Budget cuts undermine AI research and faculty hiring progress
By
–
He cut budgets.
Result: fewer faculty hires, fewer PhD students, less research, slower progress, lost technology leadership. -
Research Funding Cuts Impact Universities and PhD Programs
By
–
Not cutting the budgets of research funding agencies (NIH, NSF, etc) would be a good start to correct one of your many "carelessly ignorant, cruel," stupid, and short-sighted mistakes. Universities are rescinding offers to new faculty and PhD students right now.
-
JEPA Planning in Latent Space with Variance-Covariance Regularization
By
–
New paper: Planning in latent space with a JEPA model trained with Variance-Covariance regularization. https://t.co/tVHQUKaaVa
— Yann LeCun (@ylecun) 24 février 2025New paper: Planning in latent space with a JEPA model trained with Variance-Covariance regularization.
-
Europe’s Tech Industry: Immediate Actions for Growth
By
–
Hey Europe, you want a vibrant tech industry, right?
Do this immediately (it's genius!) -
Self-Supervised Learning: Automatic vs Human-Labeled Similarity Graphs
By
–
Correct.
You need a similarity-dissimilarity graph between samples.
If the graph is automatically constructed (e.g. through data augmentation, multiview, corruption/masking, etc), it's SSL.
If the graph uses human-provided labels, it's supervised. -
Contrastive Learning Siamese Networks Historical Development
By
–
Contrastive learning for Joint Embedding Architectures (or Siamese networks) was introduced in 1993, and again in 2005 and 2006.
– Signature verification using a" siamese" time delay neural network
Jane Bromley, Isabelle Guyon, Yann LeCun, Eduard Säckinger, Roopak Shah
