arXiv subjects

Yihan Zhang

Publications and source records attributed to Yihan Zhang.

At least 55 records · Page 3Linked to original sources

arXiv2024

Matrix Denoising with Doubly Heteroscedastic Noise: Fundamental Limits and Optimal Spectral Methods

We study the matrix denoising problem of estimating the singular vectors of a rank-$1$ signal corrupted by noise with both column and row correlations. Existing works are either unable to pinpoint the exact asymptotic estimation error or, when they do so, the resulting approaches (e.g., based on whitening or singular value shrinkage) remain vastly suboptimal. On top of this, most of the literature has focused on the special case of estimating the left singular vector of the signal when the noise only possesses row correlation (one-sided heteroscedasticity). In contrast, our work establishes the information-theoretic and algorithmic limits of matrix denoising with doubly heteroscedastic noise. We characterize the exact asymptotic minimum mean square error, and design a novel spectral estimator with rigorous optimality guarantees: under a technical condition, it attains positive correlation with the signals whenever information-theoretically possible and, for one-sided heteroscedasticity, it also achieves the Bayes-optimal error. Numerical experiments demonstrate the significant advantage of our theoretically principled method with the state of the art. The proofs draw connections with statistical physics and approximate message passing, departing drastically from standard random matrix theory techniques.

math.ST↗

arXiv2024

Quantitative Evaluation of the Saliency Map for Alzheimer's Disease Classifier with Anatomical Segmentation

Saliency maps have been widely used to interpret deep learning classifiers for Alzheimer's disease (AD). However, since AD is heterogeneous and has multiple subtypes, the pathological mechanism of AD remains not fully understood and may vary from patient to patient. Due to the lack of such understanding, it is difficult to comprehensively and effectively assess the saliency map of AD classifier. In this paper, we utilize the anatomical segmentation to allocate saliency values into different brain regions. By plotting the distributions of saliency maps corresponding to AD and NC (Normal Control), we can gain a comprehensive view of the model's decisions process. In order to leverage the fact that the brain volume shrinkage happens in AD patients during disease progression, we define a new evaluation metric, brain volume change score (VCS), by computing the average Pearson correlation of the brain volume changes and the saliency values of a model in different brain regions for each patient. Thus, the VCS metric can help us gain some knowledge of how saliency maps resulting from different models relate to the changes of the volumes across different regions in the whole brain. We trained candidate models on the ADNI dataset and tested on three different datasets. Our results indicate: (i) models with higher VCSs tend to demonstrate saliency maps with more details relevant to the AD pathology, (ii) using gradient-based adversarial training strategies such as FGSM and stochastic masking can improve the VCSs of the models.

cs.CV↗

arXiv2024

Federated Compositional Deep AUC Maximization

Federated learning has attracted increasing attention due to the promise of balancing privacy and large-scale learning; numerous approaches have been proposed. However, most existing approaches focus on problems with balanced data, and prediction performance is far from satisfactory for many real-world applications where the number of samples in different classes is highly imbalanced. To address this challenging problem, we developed a novel federated learning method for imbalanced data by directly optimizing the area under curve (AUC) score. In particular, we formulate the AUC maximization problem as a federated compositional minimax optimization problem, develop a local stochastic compositional gradient descent ascent with momentum algorithm, and provide bounds on the computational and communication complexities of our algorithm. To the best of our knowledge, this is the first work to achieve such favorable theoretical results. Finally, extensive experimental results confirm the efficacy of our method.

cs.LG↗

arXiv2024

SAT-DIFF: A Tree Diffing Framework Using SAT Solving

Computing differences between tree-structured data is a critical but challenging problem in software analysis. In this paper, we propose a novel tree diffing approach called SatDiff, which reformulates the structural diffing problem into a MaxSAT problem. By encoding the necessary transformations from the source tree to the target tree, SatDiff generates correct, minimal, and type safe low-level edit scripts with formal guarantees. We then synthesize concise high-level edit scripts by effectively merging low-level edits in the appropriate topological order. Our empirical results demonstrate that SatDiff outperforms existing heuristic-based approaches by a significant margin in terms of conciseness while maintaining a reasonable runtime.

cs.PL↗

arXiv2023

RAFT: Reward rAnked FineTuning for Generative Foundation Model Alignment

Generative foundation models are susceptible to implicit biases that can arise from extensive unsupervised training data. Such biases can produce suboptimal samples, skewed outcomes, and unfairness, with potentially serious consequences. Consequently, aligning these models with human ethics and preferences is an essential step toward ensuring their responsible and effective deployment in real-world applications. Prior research has primarily employed Reinforcement Learning from Human Feedback (RLHF) to address this problem, where generative models are fine-tuned with RL algorithms guided by a human-feedback-informed reward model. However, the inefficiencies and instabilities associated with RL algorithms frequently present substantial obstacles to the successful alignment, necessitating the development of a more robust and streamlined approach. To this end, we introduce a new framework, Reward rAnked FineTuning (RAFT), designed to align generative models effectively. Utilizing a reward model and a sufficient number of samples, our approach selects the high-quality samples, discarding those that exhibit undesired behavior, and subsequently enhancing the model by fine-tuning on these filtered samples. Our studies show that RAFT can effectively improve the model performance in both reward learning and other automated metrics in both large language models and diffusion models.

cs.LG↗

arXiv2023

Tight Bounds on List-Decodable and List-Recoverable Zero-Rate Codes

In this work, we consider the list-decodability and list-recoverability of codes in the zero-rate regime. Briefly, a code $\mathcal{C} \subseteq [q]^n$ is $(p,\ell,L)$-list-recoverable if for all tuples of input lists $(Y_1,\dots,Y_n)$ with each $Y_i \subseteq [q]$ and $|Y_i|=\ell$ the number of codewords $c \in \mathcal{C}$ such that $c_i \notin Y_i$ for at most $pn$ choices of $i \in [n]$ is less than $L$; list-decoding is the special case of $\ell=1$. In recent work by Resch, Yuan and Zhang~(ICALP~2023) the zero-rate threshold for list-recovery was determined for all parameters: that is, the work explicitly computes $p_*:=p_*(q,\ell,L)$ with the property that for all $ε>0$ (a) there exist infinite families positive-rate $(p_*-ε,\ell,L)$-list-recoverable codes, and (b) any $(p_*+ε,\ell,L)$-list-recoverable code has rate $0$. In fact, in the latter case the code has constant size, independent on $n$. However, the constant size in their work is quite large in $1/ε$, at least $|\mathcal{C}|\geq (\frac{1}ε)^{O(q^L)}$. Our contribution in this work is to show that for all choices of $q,\ell$ and $L$ with $q \geq 3$, any $(p_*+ε,\ell,L)$-list-recoverable code must have size $O_{q,\ell,L}(1/ε)$, and furthermore this upper bound is complemented by a matching lower bound $Ω_{q,\ell,L}(1/ε)$. This greatly generalizes work by Alon, Bukh and Polyanskiy~(IEEE Trans.\ Inf.\ Theory~2018) which focused only on the case of binary alphabet (and thus necessarily only list-decoding). We remark that we can in fact recover the same result for $q=2$ and even $L$, as obtained by Alon, Bukh and Polyanskiy: we thus strictly generalize their work.

cs.IT↗

arXiv2023

Codes for the Z-channel

This paper is a collection of results on combinatorial properties of codes for the Z-channel. A Z-channel with error fraction $τ$ takes as input a length-$n$ binary codeword and injects in an adversarial manner up to $nτ$ asymmetric errors, i.e., errors that only zero out bits but do not flip $0$'s to $1$'s. It is known that the largest $(L-1)$-list-decodable code for the Z-channel with error fraction $τ$ has exponential size (in $n$) if $τ$ is less than a critical value that we call the $(L-1)$-list-decoding Plotkin point and has constant size if $τ$ is larger than the threshold. The $(L-1)$-list-decoding Plotkin point is known to be $ L^{-\frac{1}{L-1}} - L^{-\frac{L}{L-1}} $, which equals $1/4$ for unique-decoding with $ L-1=1 $. In this paper, we derive various results for the size of the largest codes above and below the list-decoding Plotkin point. In particular, we show that the largest $(L-1)$-list-decodable code $ε$-above the Plotkin point, {for any given sufficiently small positive constant $ ε>0 $,} has size $Θ_L(ε^{-3/2})$ for any $L-1\ge1$. We also devise upper and lower bounds on the exponential size of codes below the list-decoding Plotkin point.

cs.IT↗

arXiv2023

Can ChatGPT Pass An Introductory Level Functional Language Programming Course?

The recent introduction of ChatGPT has drawn significant attention from both industry and academia due to its impressive capabilities in solving a diverse range of tasks, including language translation, text summarization, and computer programming. Its capability for writing, modifying, and even correcting code together with its ease of use and access is already dramatically impacting computer science education. This paper aims to explore how well ChatGPT can perform in an introductory-level functional language programming course. In our systematic evaluation, we treated ChatGPT as one of our students and demonstrated that it can achieve a grade B- and its rank in the class is 155 out of 314 students overall. Our comprehensive evaluation provides valuable insights into ChatGPT's impact from both student and instructor perspectives. Additionally, we identify several potential benefits that ChatGPT can offer to both groups. Overall, we believe that this study significantly clarifies and advances our understanding of ChatGPT's capabilities and potential impact on computer science education.

cs.CY↗

arXiv2023

Can Decentralized Stochastic Minimax Optimization Algorithms Converge Linearly for Finite-Sum Nonconvex-Nonconcave Problems?

Decentralized minimax optimization has been actively studied in the past few years due to its application in a wide range of machine learning models. However, the current theoretical understanding of its convergence rate is far from satisfactory since existing works only focus on the nonconvex-strongly-concave problem. This motivates us to study decentralized minimax optimization algorithms for the nonconvex-nonconcave problem. To this end, we develop two novel decentralized stochastic variance-reduced gradient descent ascent algorithms for the finite-sum nonconvex-nonconcave problem that satisfies the Polyak-Łojasiewicz (PL) condition. In particular, our theoretical analyses demonstrate how to conduct local updates and perform communication to achieve the linear convergence rate. To the best of our knowledge, this is the first work achieving linear convergence rates for decentralized nonconvex-nonconcave problems. Finally, we verify the performance of our algorithms on both synthetic and real-world datasets. The experimental results confirm the efficacy of our algorithms.

cs.LG↗

arXiv2022

Bounds for Multiple Packing and List-Decoding Error Exponents

We revisit the problem of high-dimensional multiple packing in Euclidean space. Multiple packing is a natural generalization of sphere packing and is defined as follows. Let $ N>0 $ and $ L\in\mathbb{Z}_{\ge2} $. A multiple packing is a set $\mathcal{C}$ of points in $ \mathbb{R}^n $ such that any point in $ \mathbb{R}^n $ lies in the intersection of at most $ L-1 $ balls of radius $ \sqrt{nN} $ around points in $ \mathcal{C} $. We study the multiple packing problem for both bounded point sets whose points have norm at most $\sqrt{nP}$ for some constant $P>0$ and unbounded point sets whose points are allowed to be anywhere in $ \mathbb{R}^n $. Given a well-known connection with coding theory, multiple packings can be viewed as the Euclidean analog of list-decodable codes, which are well-studied for finite fields. In this paper, we derive various bounds on the largest possible density of a multiple packing in both bounded and unbounded settings. A related notion called average-radius multiple packing is also studied. Some of our lower bounds exactly pin down the asymptotics of certain ensembles of average-radius list-decodable codes, e.g., (expurgated) Gaussian codes and (expurgated) Poisson Point Processes. To this end, we apply tools from high-dimensional geometry and large deviation theory. Some of our lower bounds on the optimal multiple packing density are the best known lower bounds. These bounds are obtained via a proxy known as error exponent. The latter quantity is the best exponent of the probability of list-decoding error when the code is corrupted by a Gaussian noise. We establish a curious inequality which relates the error exponent, a quantity of average-case nature, to the list-decoding radius, a quantity of worst-case nature. We derive various bounds on the error exponent in both bounded and unbounded settings which are of independent interest beyond multiple packing.

math.MG↗

arXiv2022

Multiple Packing: Lower and Upper Bounds

We study the problem of high-dimensional multiple packing in Euclidean space. Multiple packing is a natural generalization of sphere packing and is defined as follows. Let $ N>0 $ and $ L\in\mathbb{Z}_{\ge2} $. A multiple packing is a set $\mathcal{C}$ of points in $ \mathbb{R}^n $ such that any point in $ \mathbb{R}^n $ lies in the intersection of at most $ L-1 $ balls of radius $ \sqrt{nN} $ around points in $ \mathcal{C} $. We study the multiple packing problem for both bounded point sets whose points have norm at most $\sqrt{nP}$ for some constant $P>0$ and unbounded point sets whose points are allowed to be anywhere in $ \mathbb{R}^n $. Given a well-known connection with coding theory, multiple packings can be viewed as the Euclidean analog of list-decodable codes, which are well-studied for finite fields. In this paper, we derive various bounds on the largest possible density of a multiple packing in both bounded and unbounded settings. A related notion called average-radius multiple packing is also studied. Some of our lower bounds exactly pin down the asymptotics of certain ensembles of average-radius list-decodable codes, e.g., (expurgated) Gaussian codes and (expurgated) spherical codes. In particular, our lower bound obtained from spherical codes is the best known lower bound on the optimal multiple packing density and is the first lower bound that approaches the known large $L$ limit under the average-radius notion of multiple packing. To derive these results, we apply tools from high-dimensional geometry and large deviation theory.

math.MG↗

arXiv2022

Multiple Packing: Lower Bounds via Infinite Constellations

math.MG↗

arXiv2022

Multiple Packing: Lower Bounds via Error Exponents

We derive lower bounds on the maximal rates for multiple packings in high-dimensional Euclidean spaces. Multiple packing is a natural generalization of the sphere packing problem. For any $ N>0 $ and $ L\in\mathbb{Z}_{\ge2} $, a multiple packing is a set $\mathcal{C}$ of points in $ \mathbb{R}^n $ such that any point in $ \mathbb{R}^n $ lies in the intersection of at most $ L-1 $ balls of radius $ \sqrt{nN} $ around points in $ \mathcal{C} $. We study this problem for both bounded point sets whose points have norm at most $\sqrt{nP}$ for some constant $P>0$ and unbounded point sets whose points are allowed to be anywhere in $ \mathbb{R}^n $. Given a well-known connection with coding theory, multiple packings can be viewed as the Euclidean analog of list-decodable codes, which are well-studied for finite fields. We derive the best known lower bounds on the optimal multiple packing density. This is accomplished by establishing a curious inequality which relates the list-decoding error exponent for additive white Gaussian noise channels, a quantity of average-case nature, to the list-decoding radius, a quantity of worst-case nature. We also derive various bounds on the list-decoding error exponent in both bounded and unbounded settings which are of independent interest beyond multiple packing.

math.MG↗

arXiv2022

Zero-Rate Thresholds and New Capacity Bounds for List-Decoding and List-Recovery

In this work we consider the list-decodability and list-recoverability of arbitrary $q$-ary codes, for all integer values of $q\geq 2$. A code is called $(p,L)_q$-list-decodable if every radius $pn$ Hamming ball contains less than $L$ codewords; $(p,\ell,L)_q$-list-recoverability is a generalization where we place radius $pn$ Hamming balls on every point of a combinatorial rectangle with side length $\ell$ and again stipulate that there be less than $L$ codewords. Our main contribution is to precisely calculate the maximum value of $p$ for which there exist infinite families of positive rate $(p,\ell,L)_q$-list-recoverable codes, the quantity we call the zero-rate threshold. Denoting this value by $p_*$, we in fact show that codes correcting a $p_*+\varepsilon$ fraction of errors must have size $O_{\varepsilon}(1)$, i.e., independent of $n$. Such a result is typically referred to as a ``Plotkin bound.'' To complement this, a standard random code with expurgation construction shows that there exist positive rate codes correcting a $p_*-\varepsilon$ fraction of errors. We also follow a classical proof template (typically attributed to Elias and Bassalygo) to derive from the zero-rate threshold other tradeoffs between rate and decoding radius for list-decoding and list-recovery. Technically, proving the Plotkin bound boils down to demonstrating the Schur convexity of a certain function defined on the $q$-simplex as well as the convexity of a univariate function derived from it. We remark that an earlier argument claimed similar results for $q$-ary list-decoding; however, we point out that this earlier proof is flawed.

cs.IT↗

arXiv2022

Mean Estimation in High-Dimensional Binary Markov Gaussian Mixture Models

We consider a high-dimensional mean estimation problem over a binary hidden Markov model, which illuminates the interplay between memory in data, sample size, dimension, and signal strength in statistical inference. In this model, an estimator observes $n$ samples of a $d$-dimensional parameter vector $θ_{*}\in\mathbb{R}^{d}$, multiplied by a random sign $ S_i $ ($1\le i\le n$), and corrupted by isotropic standard Gaussian noise. The sequence of signs $\{S_{i}\}_{i\in[n]}\in\{-1,1\}^{n}$ is drawn from a stationary homogeneous Markov chain with flip probability $δ\in[0,1/2]$. As $δ$ varies, this model smoothly interpolates two well-studied models: the Gaussian Location Model for which $δ=0$ and the Gaussian Mixture Model for which $δ=1/2$. Assuming that the estimator knows $δ$, we establish a nearly minimax optimal (up to logarithmic factors) estimation error rate, as a function of $\|θ_{*}\|,δ,d,n$. We then provide an upper bound to the case of estimating $δ$, assuming a (possibly inaccurate) knowledge of $θ_{*}$. The bound is proved to be tight when $θ_{*}$ is an accurately known constant. These results are then combined to an algorithm which estimates $θ_{*}$ with $δ$ unknown a priori, and theoretical guarantees on its error are stated.

math.ST↗

arXiv2022

Environment Sensing Considering the Occlusion Effect: A Multi-View Approach

In this paper, we consider the problem of sensing the environment within a wireless cellular framework. Specifically, multiple user equipments (UEs) send sounding signals to one or multiple base stations (BSs) and then a centralized processor retrieves the environmental information from all the channel information obtained at the BS(s). Taking into account the occlusion effect that is common in the wireless context, we make full use of the different views of the environment from different users and/or BS(s), and propose an effective sensing algorithm called GAMP-MVSVR (generalized-approximate-message-passing-based multi-view sparse vector reconstruction). In the proposed algorithm, a multi-layer factor graph is constructed to iteratively estimate the scattering coefficients of the cloud points and their occlusion relationship. In each iteration, the occlusion relationship between the cloud points of the sparse environment is recalculated according to a simple occlusion detection rule, and in turn, used to estimate the scattering coefficients of the cloud points. Our proposed algorithm can achieve improved sensing performance with multi-BS collaboration in addition to the multi-views from the UEs. The simulation results verify its convergence and effectiveness.

eess.SP↗

arXiv2022

The Capacity of Causal Adversarial Channels

We characterize the capacity for the discrete-time arbitrarily varying channel with discrete inputs, outputs, and states when (a) the encoder and decoder do not share common randomness, (b) the input and state are subject to cost constraints, (c) the transition matrix of the channel is deterministic given the state, and (d) at each time step the adversary can only observe the current and past channel inputs when choosing the state at that time. The achievable strategy involves stochastic encoding together with list decoding and a disambiguation step. The converse uses a two-phase "babble-and-push" strategy where the adversary chooses the state randomly in the first phase, list decodes the output, and then chooses state inputs to symmetrize the channel in the second phase. These results generalize prior work on specific channels models (additive, erasure) to general discrete alphabets and models.

cs.IT↗

arXiv2021

List Decoding Random Euclidean Codes and Infinite Constellations

We study the list decodability of different ensembles of codes over the real alphabet under the assumption of an omniscient adversary. It is a well-known result that when the source and the adversary have power constraints $ P $ and $ N $ respectively, the list decoding capacity is equal to $ \frac{1}{2}\log\frac{P}{N} $. Random spherical codes achieve constant list sizes, and the goal of the present paper is to obtain a better understanding of the smallest achievable list size as a function of the gap to capacity. We show a reduction from arbitrary codes to spherical codes, and derive a lower bound on the list size of typical random spherical codes. We also give an upper bound on the list size achievable using nested Construction-A lattices and infinite Construction-A lattices. We then define and study a class of infinite constellations that generalize Construction-A lattices and prove upper and lower bounds for the same. Other goodness properties such as packing goodness and AWGN goodness of infinite constellations are proved along the way. Finally, we consider random lattices sampled from the Haar distribution and show that if a certain number-theoretic conjecture is true, then the list size grows as a polynomial function of the gap-to-capacity.

cs.IT↗