~/talk

Towards Verification of General Neural Networks: A Dual Approach

Talkat Real-Time Decision Making 2018 · 2018-04-30..05-04· [machine-learning]
links:none on file
slides:none on file

Poses verification as an optimization and uses a dual Lagrangian relaxation to upper‑bound worst‑case adversarial scores; proves NP‑hardness for even one hidden layer but gives exact/approximate guarantees for two‑layer nets and tight bounds on MNIST up to epsilon ≈ 30 (about 10% grayscale). Implements a learned verifier network to predict dual variables, folding verification into training; same architecture yields better verified accuracy and drops wall‑clock training from ~5 hours to ~20 minutes.

AI-written summary · about

  1. 0:00Start
  2. 2:52Specifying tasks
  3. 3:35What is supervised learning? Training Data
  4. 5:52Algorithmic bias and fairness
  5. 6:20We need richer specifications for ML models
  6. 9:56Defenses against adversarial examples
  7. 11:01Worst-case guarantees: Verification Module level verification
  8. 13:23Hardness of verification in general
  9. 15:35Comparison to prior work on verification of NNS
  10. 21:29Verification as optimization
  11. 23:06Layers as constraints
  12. 24:41Lagrangian relaxation
  13. 26:30Verification using duality: advantages
  14. 27:43Results: Adversarial attacks
  15. 29:30Results: Classifier stability
  16. 31:25Learning verifiers
  17. 32:40Open questions and challenges Tractable verification: Under what conditions can verification be done tractably? Results for single hidden layer networks. can be extended beyond?
  18. 33:16Open questions and challenges Tractable verification: Under what conditions can verification be done tractably? Results for single hidden layer networks - can be extended beyond?
~/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.