New framework improves text watermark detection under imperfect pseudorandomness.
problem Structured dependence in generated text from language models causes Type I error control issues.
method Hierarchical two-layer partition, minimal units, non-asymptotic efficiency measure, minimax hypothesis testing.
result Closed-form optimal rules for watermark detection under imperfect pseudorandomness.
Paper proves hardness of learning various complex models under local pseudorandom generators.
problem Hardness of learning various complex models.
method Existence of local pseudorandom generators.
result Proves hardness of learning shallow ReLU neural networks and other models.
Generation of pseudorandom numbers from different probability distributions has been studied extensively in the Monte Carlo simulation literature. Two standard generation techniques are the acceptance-rejection and inverse transformation methods. An alternative approach to Monte Carlo simulation is the quasi-Monte Carl…
Arora, Barak, Brunnermeier, and Ge showed that taking computational complexity into account, a dishonest seller could strategically place lemons in financial derivatives to make them substantially less valuable to buyers. We show that if the seller is required to construct derivatives of a certain form, then this pheno…
New findings show learning deeper neural networks is hard even with Gaussian inputs and non-degenerate weights.
problem The computational complexity of learning neural networks, especially deeper ones.
method Smoothed analysis framework and local pseudorandom generators.
result Learning depth-3 ReLU networks under Gaussian input distribution is hard even if weight matrices are non-degenerate.
Quantum speedup for Monte Carlo integration reduces integrand calls.
problem Reducing the number of calls to the integrand subroutine in high-dimensional Monte Carlo integration.
method Combining nested quantum amplitude estimation with pseudorandom numbers for separable integrands.
result Significant reduction in the number of integrand calls for high-dimensional integration.
Paper proves first non-trivial PTF testing lower bounds for NGCA.
problem Proving lower bounds against PTF tests is challenging.
method Developed tools to prove PTF testing lower bounds for NGCA.
result First non-trivial PTF testing lower bounds for NGCA.
Power laws detected in financial data, modeled with random multipliers.
problem Detecting power laws in financial data.
method Investigated data from financial instruments, proposed a model based on sums of Maxwell-Boltzmann distributions with random multipliers.
result Detected power laws with various exponents in financial data, proposed a universal model.
We consider the problem of simulating loss probabilities and conditional excesses for linear asset portfolios under the t-copula model. Although in the literature on market risk management there are papers proposing efficient variance reduction methods for Monte Carlo simulation of portfolio market risk, there is no pa…
The paper shows how shared random seeds can reduce variance in machine learning evaluations.
problem The statistical structure of comparative evaluation under shared random seeds is not well understood.
method An extended learning-based multi-agent economic simulator was used to demonstrate the effects of shared random seeds on variance reduction.
result Pairing seeds can reduce variance in machine learning evaluations, especially when outcomes are positively correlated at the seed level.
A new watermarking method corrects bias in language models using maximal coupling.
problem Correcting bias in language model token distributions.
method Maximal coupling to balance bias correction and text quality.
result Outperforms prior techniques in preserving text quality and detectability.
Next-gen reservoir computing models dynamical systems from time-series data.
problem Modeling dynamical systems from time-series data.
method Pseudorandom nonlinear projection of time-delay embedded inputs.
result Models remain stable over long rollouts and generalize beyond training data.
Estimates proportions of LLM-generated text in mixed documents.
problem Estimating the proportion of text generated by a pre-specified LLM in mixed documents.
method Developed estimators for two observation regimes: full observation and pivotal reduction, and established sample complexity bounds.
result Full observation estimators require fewer samples than pivotal reduction estimators.
New findings suggest minimax optimality doesn't guarantee distribution learning for GANs.
problem Understanding when GANs can truly learn the underlying distribution.
method Using cryptographic assumptions and ReLU network generators, the paper shows that achieving minimax optimality is insufficient for distribution learning.
result Achieving minimax optimality is insufficient for distribution learning in the usual statistical sense.
Study on estimating Gumbel--Max watermark proportions in edited documents.
problem Estimating the proportion of a document generated from a watermarked LLM.
method Comparison of full observation and pivotal reduction observation regimes; development of estimators and information-theoretic lower bounds.
result Full observation yields a substantially smaller sample complexity compared to pivotal reduction.
New framework to understand and exploit curvature in deep learning loss landscapes.
problem Understanding and optimizing the loss landscape in deep learning models.
method New conceptual framework and techniques to estimate and exploit curvature of expected loss changes.
result Alice algorithm optimizes training by incorporating curvature terms and step bounds.
New concept of epiplexity quantifies useful information from data.
problem Understanding useful information content from data without unlimited computational capacity.
method Introducing epiplexity, a measure of information computationally bounded observers can learn.
result Epiplexity captures useful information content, not just randomness.