Search Results for author:
Found 59 papers, 28 papers with code
Evaluating the Unseen Capabilities: How Many Theorems Do LLMs Know?
Accurate evaluation of large language models (LLMs) is crucial for understanding their capabilities and guiding their development.
Fundamental Limits of Game-Theoretic LLM Alignment: Smith Consistency and Preference Matching
However, using raw preference as the payoff in the game highly limits the potential of the game-theoretic LLM alignment framework.
Restoring Calibration for Aligned Large Language Models: A Calibration-Aware Fine-Tuning Approach
However, as models are further fine-tuned for better performance, they enter the non-calibratable regime.
Statistical Impossibility and Possibility of Aligning LLMs with Human Preferences: From Condorcet Paradox to Nash Equilibrium
As a blessing, we prove that this condition holds with high probability under the probabilistic preference model, thereby highlighting the statistical possibility of preserving minority preferences without explicit regularization in aligning LLMs.
An Overview of Large Language Models for Statisticians
Large Language Models (LLMs) have emerged as transformative tools in artificial intelligence (AI), exhibiting remarkable capabilities across diverse tasks such as text generation, reasoning, and decision-making.
Robust Detection of Watermarks for Large Language Models Under Human Edits
We prove that the Tr-GoF test achieves optimality in robust detection of the Gumbel-max watermark in a certain asymptotic regime of substantial text modifications and vanishing watermark signals.
Debiasing Watermarks for Large Language Models via Maximal Coupling
Watermarking language models is essential for distinguishing between human and machine-generated text and thus maintaining the integrity and trustworthiness of digital communication.
Magnetic Preference Optimization: Achieving Last-iterate Convergence for Language Model Alignment
Self-play methods have demonstrated remarkable success in enhancing model capabilities across various domains.
The 2020 United States Decennial Census Is More Private Than You (Might) Think
The U. S. Decennial Census serves as the foundation for many high-profile policy decision-making processes, including federal funding allocation and redistricting.
A Statistical Viewpoint on Differential Privacy: Hypothesis Testing, Representation and Blackwell's Theorem
We review techniques that render $f$-differential privacy a unified framework for analyzing privacy bounds in data analysis and machine learning.
A Law of Next-Token Prediction in Large Language Models
Large language models (LLMs) have been widely employed across various application domains, yet their black-box nature poses significant challenges to understanding how these models process input data internally to make predictions.
A Peek into Token Bias: Large Language Models Are Not Yet Genuine Reasoners
This study introduces a hypothesis-testing framework to assess whether large language models (LLMs) possess genuine reasoning abilities or primarily depend on token bias.
Bridging the Gap: Rademacher Complexity in Robust and Standard Generalization
We aim to construct a new cover that possesses two properties: 1) compatibility with adversarial examples, and 2) precision comparable to covers used in standard settings.
Tackling Copyright Issues in AI Image Generation Through Originality Estimation and Genericization
As a practical implementation, we introduce PREGen (Prompt Rewriting-Enhanced Genericization), which combines our genericization method with an existing mitigation technique.
Towards Rationality in Language and Multimodal Agents: A Survey
This work discusses how to build more rational language and multimodal agents and what criteria define rationality in intelligent systems.
On the Algorithmic Bias of Aligning Large Language Models with RLHF: Preference Collapse and Matching Regularization
To mitigate this algorithmic bias, we introduce preference matching (PM) RLHF, a novel approach that provably aligns LLMs with the preference distribution of the reward model under the Bradley--Terry--Luce/Plackett--Luce model.
Neural Collapse Meets Differential Privacy: Curious Behaviors of NoisyGD with Near-perfect Representation Learning
A recent study by De et al. (2022) has reported that large-scale representation learning through pre-training on a public dataset significantly enhances differentially private (DP) learning in downstream tasks, despite the high dimensionality of the feature space.
An Economic Solution to Copyright Challenges of Generative AI
Generative artificial intelligence (AI) systems are trained on large data corpora to generate new pieces of text, images, videos, and other media.
A Statistical Framework of Watermarks for Large Language Models: Pivot, Detection Efficiency and Optimal Rules
In particular, we derive optimal detection rules for these watermarks under our framework.
Provable Multi-Party Reinforcement Learning with Diverse Human Feedback
Our work \textit{initiates} the theoretical study of multi-party RLHF that explicitly models the diverse preferences of multiple individuals.
What Should Data Science Education Do with Large Language Models?
LLMs can also play a significant role in the classroom as interactive teaching and learning tools, contributing to personalized education.
DP-HyPO: An Adaptive Private Hyperparameter Optimization Framework
In our paper, we introduce DP-HyPO, a pioneering framework for ``adaptive'' private hyperparameter optimization, aiming to bridge the gap between private and non-private hyperparameter optimization.
Reward Collapse in Aligning Large Language Models
This insight allows us to derive closed-form expressions for the reward distribution associated with a set of utility functions in an asymptotic regime.
The Implicit Regularization of Dynamical Stability in Stochastic Gradient Descent
By contrast, for gradient descent (GD), the stability imposes a similar constraint but only on the largest eigenvalue of Hessian.
Isotonic Mechanism for Exponential Family Estimation in Machine Learning Peer Review
In 2023, the International Conference on Machine Learning (ICML) required authors with multiple submissions to rank their submissions based on perceived quality.
A Law of Data Separation in Deep Learning
While deep learning has enabled significant advances in many areas of science, its black-box nature hinders architecture design for future artificial intelligence applications and interpretation for high-stakes decision makings.
On Quantum Speedups for Nonconvex Optimization via Quantum Tunneling Walks
Classical algorithms are often not effective for solving nonconvex optimization problems where local minima are separated by high barriers.
Analytical Composition of Differential Privacy via the Edgeworth Accountant
Many modern machine learning algorithms are composed of simple private algorithms; thus, an increasingly important problem is to efficiently compute the overall privacy loss under composition.
FIFA: Making Fairness More Generalizable in Classifiers Trained on Imbalanced Data
Specifically, FIFA encourages both classification and fairness generalization and can be flexibly combined with many existing fair learning methods with logits-based losses.
ROCK: Causal Inference Principles for Reasoning about Commonsense Causality
Commonsense causality reasoning (CCR) aims at identifying plausible causes and effects in natural language descriptions that are deemed reasonable by an average person.
Envisioning Future Deep Learning Theories: Some Basic Concepts and Characteristics
To advance deep learning methodologies in the next decade, a theoretical framework for reasoning about modern neural networks is needed.
You Are the Best Reviewer of Your Own Papers: An Owner-Assisted Scoring Mechanism
To address this withholding of information, in this paper, I introduce the Isotonic Mechanism, a simple and efficient approach to improving imprecise raw scores by leveraging certain information that the owner is incentivized to provide.
Imitating Deep Learning Dynamics via Locally Elastic Stochastic Differential Equations
Our main finding uncovers a sharp phase transition phenomenon regarding the {intra-class impact: if the SDEs are locally elastic in the sense that the impact is more significant on samples from the same class as the input, the features of the training data become linearly separable, meaning vanishing training loss; otherwise, the features are not separable, regardless of how long the training time is.
An Unconstrained Layer-Peeled Perspective on Neural Collapse
We prove that gradient flow on this model converges to critical points of a minimum-norm separation problem exhibiting neural collapse in its global minimizer.
Weighted Training for Cross-Task Learning
In this paper, we introduce Target-Aware Weighted Training (TAWT), a weighted training algorithm for cross-task learning based on minimizing a representation-based task distance between the source and target tasks.
Characterizing the SLOPE Trade-off: A Variational Perspective and the Donoho-Tanner Limit
Sorted l1 regularization has been incorporated into many methods for solving high-dimensional statistical estimation problems, including the SLOPE estimator in linear regression.
Oneshot Differentially Private Top-k Selection
Being able to efficiently and accurately select the top-$k$ elements with differential privacy is an integral component of various private data analysis tasks.
Rejoinder: Gaussian Differential Privacy
In this rejoinder, we aim to address two broad issues that cover most comments made in the discussion.
A Central Limit Theorem for Differentially Private Query Answering
The central question, therefore, is to understand which noise distribution optimizes the privacy-accuracy trade-off, especially when the dimension of the answer vector is high.
A Theorem of the Alternative for Personalized Federated Learning
A widely recognized difficulty in federated learning arises from the statistical heterogeneity among clients: local datasets often come from different but not entirely unrelated distributions, and personalization is, therefore, necessary to achieve optimal results from each individual's perspective.
Federated $f$-Differential Privacy
Federated learning (FL) is a training paradigm where the clients collaboratively learn models by repeatedly sharing information without compromising much on the privacy of their local sensitive data.
Exploring Deep Neural Networks via Layer-Peeled Model: Minority Collapse in Imbalanced Training
More importantly, when moving to the imbalanced case, our analysis of the Layer-Peeled Model reveals a hitherto unknown phenomenon that we term \textit{Minority Collapse}, which fundamentally limits the performance of deep learning models on the minority classes.
Toward Better Generalization Bounds with Locally Elastic Stability
Given that, we propose \emph{locally elastic stability} as a weaker and distribution-dependent stability notion, which still yields exponential generalization bounds.
Precise High-Dimensional Asymptotics for Quantifying Heterogeneous Transfers
For example, we can identify a phase transition in the high-dimensional linear regression setting from positive transfer to negative transfer under a model shift between the source and target tasks.
Label-Aware Neural Tangent Kernel: Toward Better Generalization and Local Elasticity
As a popular approach to modeling the dynamics of training overparametrized neural networks (NNs), the neural tangent kernels (NTK) are known to fall behind real-world NNs in generalization ability.
Towards Understanding the Dynamics of the First-Order Adversaries
An acknowledged weakness of neural networks is their vulnerability to adversarial perturbations to the inputs.
The Complete Lasso Tradeoff Diagram
A fundamental problem in the high-dimensional regression is to understand the tradeoff between type I and type II errors or, equivalently, false discovery rate (FDR) and power in variable selection.
Statistics Theory Information Theory Information Theory Statistics Theory
On Learning Rates and Schrödinger Operators
In this paper, we present a general theoretical analysis of the effect of the learning rate in stochastic gradient descent (SGD).
Sharp Composition Bounds for Gaussian Differential Privacy via Edgeworth Expansion
To address this question, we introduce a family of analytical and sharp privacy bounds under composition using the Edgeworth expansion in the framework of the recently proposed f-differential privacy.
Deep Learning with Gaussian Differential Privacy
Leveraging the appealing properties of $f$-differential privacy in handling composition and subsampling, this paper derives analytically tractable expressions for the privacy guarantees of both stochastic gradient descent and Adam used in training deep neural networks, without the need of developing sophisticated techniques as [3] did.
The Local Elasticity of Neural Networks
This phenomenon is shown to persist for neural networks with nonlinear activation functions through extensive simulations on real-life and synthetic datasets, whereas this is not observed in linear classifiers.
Gaussian Differential Privacy
More precisely, the privacy guarantees of \emph{any} hypothesis testing based definition of privacy (including original DP) converges to GDP in the limit under composition.
Acceleration via Symplectic Discretization of High-Resolution Differential Equations
We study first-order optimization methods obtained by discretizing ordinary differential equations (ODEs) corresponding to Nesterov's accelerated gradient methods (NAGs) and Polyak's heavy-ball method.
Understanding the Acceleration Phenomenon via High-Resolution Differential Equations
We also show that these ODEs are more accurate surrogates for the underlying algorithms; in particular, they not only distinguish between NAG-SC and Polyak's heavy-ball method, but they allow the identification of a term that we refer to as "gradient correction" that is present in NAG-SC but not in the heavy-ball method and is responsible for the qualitative difference in convergence of the two methods.
Differentially Private False Discovery Rate Control
Differential privacy provides a rigorous framework for privacy-preserving data analysis.
Robust Inference Under Heteroskedasticity via the Hadamard Estimator
In this paper, we propose methods that are robust to large and unequal noise in different observational units (i. e., heteroskedasticity) for statistical inference in linear regression.
Statistics Theory Methodology Statistics Theory
HiGrad: Uncertainty Quantification for Online Learning and Stochastic Approximation
Stochastic gradient descent (SGD) is an immensely popular approach for online learning in settings where data arrives in a stream or data sizes are very large.
When Is the First Spurious Variable Selected by Sequential Regression Procedures?
In a regime of certain sparsity levels, however, three examples of sequential procedures--forward stepwise, the lasso, and least angle regression--are shown to include the first spurious variable unexpectedly early.
Detecting Multiple Replicating Signals using Adaptive Filtering Procedures
Replicability is a fundamental quality of scientific discoveries: we are interested in those signals that are detectable in different laboratories, study populations, across time etc.
Methodology
