arXiv subjects

Xuanye Wang

Publications and source records attributed to Xuanye Wang.

4 recordsLinked to original sources

arXiv2026

No Time to Collapse: Unlocking Robustness and Multiplexed Capacity in Frozen Audio Watermarkers

Modern neural audio watermarking systems typically embed a message repeatedly across time and then collapse the resulting temporal evidence into a single payload using averaging, voting, or another fixed aggregation rule. We argue that this temporal collapse limits both robustness and the recovery of multiple payloads, and that the limitation can be addressed without retraining the underlying watermarker. We freeze a pretrained watermarker's encoder and detector and train only a low-latency Conformer-based decoder. The decoder consumes the detector's temporal soft outputs, which a system-specific adapter pools into a sequence of window-level representations, and predicts the embedded message. On three frozen watermarkers (AURA, AudioSeal, and WavMark), the learned decoder improves recovery of attacked messages and yields higher detection AUROC point estimates on all three. Under controlled full- and partial-coverage multiplexing, it improves joint-exact recovery of two alternating payload words by 9.7-48.0, 6.9-17.3, and 8.2-14.2 percentage points, respectively, under one to three chained attacks on feasible clips.

cs.SD↗

arXiv2024

Long run consequence of p-hacking

We study the theoretical consequence of p-hacking on the accumulation of knowledge under the framework of mis-specified Bayesian learning. A sequence of researchers, in turn, choose projects that generate noisy information in a field. In choosing projects, researchers need to carefully balance as projects generates big information are less likely to succeed. In doing the project, a researcher p-hacks at intensity $\varepsilon$ so that the success probability of a chosen project increases (unduly) by a constant $\varepsilon$. In interpreting previous results, researcher behaves as if there is no p-hacking because the intensity $\varepsilon$ is unknown and presumably small. We show that over-incentivizing information provision leads to the failure of learning as long as $\varepsilon\neq 0$. If the incentives of information provision is properly provided, learning is correct almost surely as long as $\varepsilon$ is small.

econ.TH↗

arXiv2021

Fragility of Confounded Learning

We consider an observational learning model with exogenous public payoff shock. We show that confounded learning doesn't arise for almost all private signals and almost all shocks, even if players have sufficiently divergent preferences.

econ.TH↗

arXiv2012

Extra structure and the universal construction for the Witten-Reshetikhin-Turaev TQFT

A TQFT is a functor from a cobordism category to the category of vector spaces, satisfying certain properties. An important property is that the vector spaces should be finite dimensional. For the WRT TQFT, the relevant 2+1-cobordism category is built from manifolds which are equipped with an extra structure such as a p_1-structure, or an extended manifold structure. We perform the universal construction of Blanchet, Habegger, Masbaum and Vogel on a cobordism category without this extra structure and show that the resulting quantization functor assigns an infinite dimensional vector space to the torus.

math.GT↗