Search for: All records

The offline reinforcement learning (RL) problem is often motivated by the need to learn data-driven decision policies in financial, legal and healthcare applications. However, the learned policy could retain sensitive information of individuals in the training data (e.g., treatment and outcome of patients), thus susceptible to various privacy risks. We design offline RL algorithms with differential privacy guarantees which provably prevent such risks. These algorithms also enjoy strong instance-dependent learning bounds under both tabular and linear Markov Decision Process (MDP) settings. Our theory and simulation suggest that the privacy guarantee comes at (almost) no drop in utility comparing to the non-private counterpart for a medium-size dataset. 

In the arena of privacy-preserving machine learning, differentially private stochastic gradient descent (DP-SGD) has outstripped the objective perturbation mechanism in popularity and interest. Though unrivaled in versatility, DP-SGD requires a non-trivial privacy overhead (for privately tuning the model’s hyperparameters) and a computational complexity which might be extravagant for simple models such as linear and logistic regression. This paper revamps the objective perturbation mechanism with tighter privacy analyses and new computational tools that boost it to perform competitively with DP-SGD on unconstrained convex generalized linear problems. 

Achieving spin-pinning at the interface of hetero-bilayer ferromagnet/antiferromagnet structures in conventional exchange bias systems can be challenging due to difficulties in interface control and the weakening of spin-pinning caused by poor interface quality. In this work, we propose an alternative approach to stabilize the exchange interaction at the interface of an uncompensated antiferromagnet by utilizing a gradient of interlayer exchange coupling. We demonstrate this exchange interaction through a designed field training protocol in the odd-layer topological antiferromagnet MnBi₂Te₄. Our results reveal a remarkable field-trained exchange bias of up to ~ 400 mT, which exhibits high repeatability and can be easily reset by a large training field. Notably, this field-trained exchange bias effect persists even with zero-field initialization, presenting a stark contrast to the traditional field-cooled exchange bias. The highly tunable exchange bias observed in this single antiferromagnet compound, without the need for an additional magnetic layer, provides valuable insight into the exchange interaction mechanism. These findings pave the way for the systematic design of topological antiferromagnetic spintronics.

 

Data valuation, a growing field that aims at quantifying the usefulness of individual data sources for training machine learning (ML) models, faces notable yet often overlooked privacy challenges. This paper studies these challenges with a focus on KNN-Shapley, one of the most practical data valuation methods nowadays. We first emphasize the inherent privacy risks of KNN-Shapley, and demonstrate the significant technical challenges in adapting KNN-Shapley to accommodate differential privacy (DP). To overcome these challenges, we introduce TKNN-Shapley, a refined variant of KNN-Shapley that is privacy-friendly, allowing for straightforward modifications to incorporate DP guarantee (DP-TKNN-Shapley). We show that DP-TKNN-Shapley has several advantages and offers a superior privacy-utility tradeoff compared to naively privatized KNN-Shapley. Moreover, even non-private TKNN-Shapley matches KNN-Shapley's performance in discerning data quality. Overall, our findings suggest that TKNN-Shapley is a promising alternative to KNN-Shapley, particularly for real-world applications involving sensitive data. 

We propose an autonomous quantum error correction scheme using squeezed cat (SC) code against excitation loss in continuous-variable systems. Through reservoir engineering, we show that a structured dissipation can stabilize a two-component SC while autonomously correcting the errors. The implementation of such dissipation only requires low-order nonlinear couplings among three bosonic modes or between a bosonic mode and a qutrit. While our proposed scheme is device independent, it is readily implementable with current experimental platforms such as superconducting circuits and trapped-ion systems. Compared to the stabilized cat, the stabilized SC has a much lower dominant error rate and a significantly enhanced noise bias. Furthermore, the bias-preserving operations for the SC have much lower error rates. In combination, the stabilized SC leads to substantially better logical performance when concatenating with an outer discrete-variable code. The surface-SC scheme achieves more than one order of magnitude increase in the threshold ratio between the loss rate κ1 and the engineered dissipation rate κ2. Under a practical noise ratio κ1/κ2 = 10−3, the repetition-SC scheme can reach a 10−15 logical error rate even with a small mean excitation number of 4, which already suffices for practically useful quantum algorithms. 

Abstract In the Asia–Pacific region (APR), extreme precipitation is one of the most critical climate stressors, affecting 60% of the population and adding pressure to governance, economic, environmental, and public health challenges. In this study, we analyzed extreme precipitation spatiotemporal trends in APR using 11 different indices and revealed the dominant factors governing precipitation amount by attributing its variability to precipitation frequency and intensity. We further investigated how these extreme precipitation indices are influenced by El Niño-Southern Oscillation (ENSO) at a seasonal scale. The analysis covered 465 ERA5 (the fifth-generation atmospheric reanalysis of the European Center for Medium-Range Weather Forecasts) study locations over eight countries and regions during 1990–2019. Results revealed a general decrease indicated by the extreme precipitation indices (e.g., the annual total amount of wet-day precipitation, average intensity of wet-day precipitation), particularly in central-eastern China, Bangladesh, eastern India, Peninsular Malaysia and Indonesia. We observed that the seasonal variability of the amount of wet-day precipitation in most locations in China and India are dominated by precipitation intensity in June–August (JJA), and by precipitation frequency in December–February (DJF). Locations in Malaysia and Indonesia are mostly dominated by precipitation intensity in March–May (MAM) and DJF. During ENSO positive phase, significant negative anomalies in seasonal precipitation indices (amount of wet-day precipitation, number of wet days and intensity of wet-day precipitation) were observed in Indonesia, while opposite results were observed for ENSO negative phase. These findings revealing patterns and drivers for extreme precipitation in APR may inform climate change adaptation and disaster risk reduction strategies in the study region. 

Computation of injective (or inversion-free) maps is a key task in geometry processing, physical simulation, and shape optimization. Despite being a longstanding problem, it remains challenging due to its highly nonconvex and combinatoric nature. We propose computation ofvariational quasi-harmonic mapsto obtain smooth inversion-free maps. Our work is built on a key observation about inversion-free maps: A planar map is a diffeomorphism if and only if it is quasi-harmonic and satisfies a special Cauchy boundary condition. We hence equate the inversion-free mapping problem to an optimal control problem derived from our theoretical result, in which we search in the space of parameters that define an elliptic PDE. We show that this problem can be solved by minimizing within a family of functionals. Similarly, our discretized functionals admit exactly injective maps as the minimizers, empirically producing inversion-free discrete maps of triangle meshes. We design efficient numerical procedures for our problem that prioritize robust convergence paths. Experiments show that on challenging examples our methods can achieve up to orders of magnitude improvement over state-of-the-art, in terms of speed or quality. Moreover, we demonstrate how to optimize a generic energy in our framework while restricting to quasi-harmonic maps.

 

We consider the problem of global optimization with noisy zeroth order oracles — a well-motivated problem useful for various applications ranging from hyper-parameter tuning for deep learning to new material design. Existing work relies on Gaussian processes or other non-parametric family, which suffers from the curse of dimensionality. In this paper, we propose a new algorithm GO-UCB that leverages a parametric family of functions (e.g., neural networks) instead. Under a realizable assumption and a few other mild geometric conditions, we show that GO-UCB achieves a cumulative regret of $\tilde{O}(\sqrt{T})$ where $T$ is the time horizon. At the core of GO-UCB is a carefully designed uncertainty set over parameters based on gradients that allows optimistic exploration. Synthetic and real-world experiments illustrate GO-UCB works better than popular Bayesian optimization approaches, even if the model is misspecified. 

Abstract Developing an eco-friendly, efficient, and highly selective gold-recovery technology is urgently needed in order to maintain sustainable environments and improve the utilization of resources. Here we report an additive-induced gold recovery paradigm based on precisely controlling the reciprocal transformation and instantaneous assembly of the second-sphere coordinated adducts formed between β-cyclodextrin and tetrabromoaurate anions. The additives initiate a rapid assembly process by co-occupying the binding cavity of β-cyclodextrin along with the tetrabromoaurate anions, leading to the formation of supramolecular polymers that precipitate from aqueous solutions as cocrystals. The efficiency of gold recovery reaches 99.8% when dibutyl carbitol is deployed as the additive. This cocrystallization is highly selective for square-planar tetrabromoaurate anions. In a laboratory-scale gold-recovery protocol, over 94% of gold in electronic waste was recovered at gold concentrations as low as 9.3 ppm. This simple protocol constitutes a promising paradigm for the sustainable recovery of gold, featuring reduced energy consumption, low cost inputs, and the avoidance of environmental pollution.