It is 2013, and a disquieting discovery is circulating: a deep network that classifies images with superhuman confidence can be flipped to a wildly wrong label by a perturbation so small no human eye can detect it. The image looks identical; the model calls a school bus an ostrich. Characterize why high-accuracy classifiers are so brittle at these engineered points, and build defenses that hold when an adversary — not nature — chooses the input. The stakes leave the lab immediately: a stop sign with a few stickers, a face-recognition gate, a malware detector. Get it wrong and you deploy a system whose benchmark accuracy is a fiction the moment someone wants it to fail — and in security, someone always does.
Chose The Bayesian network with belief propagation — wrong. The causal diagram and do-calculus was the one that fit.
Pearl's causal inference framework offers a genuinely relevant later theoretical perspective on adversarial vulnerability: a classifier relying on spurious correlations rather than causally robust features is precisely the kind of model an adversary can exploit, and causal machine learning has become a real research direction for building more robust classifiers, connecting his framework to this problem's defense-building half. He has been notably skeptical of deep learning's lack of causal structure, a critique that directly anticipates why these models are so brittle. He never worked on adversarial examples in the specific 2013 technical sense. His relevance is genuine conceptual grounding for a leading robustness research direction. Causal robustness is a real, active response to exactly this problem's failure mode.
Anandkumar's tensor methods for machine learning and later research have engaged with provable robustness guarantees for neural networks, including work connecting her tensor-decomposition expertise to certifying robustness bounds, giving her genuine technical relevance to this problem's defense-building half. Her broader research on the theoretical foundations of deep learning touches questions of model sensitivity and robustness relevant to adversarial vulnerability. She did not co-author the founding 2013 discovery paper. Her relevance is real through later technical contributions to provable robustness, secondary to the original discoverers but genuinely engaged with the defense-building side of this problem's stakes. Her provable-robustness work is a genuine, if secondary, technical contribution. Her signature contributions remain in a related but distinct theoretical corner.
Battle #103 · 8/10/2026, 11:38:01 AM · this result is deterministic: the same two personas on this problem always resolve the same way.