~/talk

Semi-Random Models and Robustness in Unsupervised Learning

links:none on file
slides:none on file

A semi-random model mixes a tiny fraction beta of truly random sparse supports with adversarially chosen supports and shows polynomial-time recovery of the dictionary A and sparse codes X up to any accuracy when incoherence holds and sparsity up to sqrt(n) (purely random analysis extends to about n^{2/3}). Recovery uses a testing procedure that reads inner-product histograms (bimodal if a candidate equals a column) plus weighted-sum candidate generation, proven via coupling/CLT and concentration bounds; identifiability follows from a mild triples condition.

AI-written summary · about

  1. 0:00Semi-random Models and Robustness in Unsupervised Learning
  2. 0:31Probabilistic Modeling in ML
  3. 1:33Semirandom models
  4. 4:19Monotone Adversaries Feige-Kilian
  5. 5:22Monotone Adversaries - 1" type **Improving" planted solution
  6. 8:06Monotone Adversaries - 2nd type Additional information about signal
  7. 15:57Guarantees for Dictionary Learning
  8. 20:56Goal
  9. 26:58Main Results
  10. 29:40o(n2/3) Sparsity in the Random Model (B = 1)
  11. 32:09Identifiability Under Minimal Assumptions
  12. 33:23Techniques
  13. 34:52Testing Procedure
  14. 38:30Key Lemma (Soundness for NO case)
  15. 40:59Candidate Generation
  16. 44:25Conclusion
~/search
type to search 6,249 conferences, 138,987 speakers and 185,172 talks
~/help
cmdctrl k
search every conference, speaker and series
/
focus the filter search
h
home
c
conferences
s
speakers
e
series
t
light / dark — the status bar button does the same
?
this help
esc
close this, then clear every filter — or just the search, while you are typing in it

Ignored while you are typing. Pages without a filter rail have no search to focus.