Second-order optimizers retain residual information after data deletion, affecting machine unlearning.
problem Residual information in second-order optimizers after data deletion.
method Comparison of first-order and second-order learners, eigendecomposition analysis.
result Second-order optimizers retain residual information, not detectable by first-order analysis.
New protocols show 1-bit mean estimation can be order-optimal without interaction.
problem Can 1-bit mean estimation be optimal without interaction?
method Adaptive and non-adaptive threshold and interval queries, with one adaptive transition.
result Arbitrary non-adaptive quantizers can match the adaptive rate, suggesting interaction is not necessary.
Data whitening and second order optimization harm generalization by reducing access to dataset information.
problem Harmful effects of data whitening and second order optimization on generalization in machine learning.
method Analysis of fully connected models and experimental verification.
result Data whitening and second order optimization reduce or prevent generalization by limiting access to dataset information.
Higher-order optimization problems naturally appear when investigating the effects of a patent with finite length, as in the pioneering work of Futagami and Iwaisako (2007). In this paper, we establish the Euler equations and transversality conditions necessary for analyzing such higher-order optimization problems. We …
Unified theory for UCB policies in total and max bandit problems.
problem Order optimality of UCB policies in max bandit problems.
method Unified definition of UCB policy using oracle quantity and failure count.
result UCB policies are order optimal in both total and max bandit problems.
OptEx accelerates first-order optimization with parallelized iterations.
problem Inefficiencies in first-order optimization algorithms for complex tasks.
method Approximately parallelized iterations using kernelized gradient estimation.
result OptEx achieves substantial efficiency improvements with an effective acceleration rate of Ω ( N ) Ω(\sqrt{N}) Ω ( N ) . CyBeR-0 optimizes federated learning with Byzantine resilience and reduced communication costs.
problem Byzantine attacks and communication inefficiency in federated learning.
method Transformed robust aggregation for zero-order optimization under client heterogeneity.
result CyBeR-0 achieves stable performance with minimal communication costs and reduced memory usage.
Second-order optimization speeds up deep hedging for complex options.
problem Hedging exotic options with market frictions in realistic markets.
method Second-order optimization scheme leveraging pathwise differentiability and Kronecker-factoring.
result Our method optimizes the policy in 1/4 the steps of standard optimization.
Transformers can approximate Newton's method for logistic regression.
problem Implementing higher order optimization methods in Transformers.
method Linear attention Transformers with ReLU layers approximating second order optimization algorithms.
result Transformers can implement a single step of Newton's iteration for matrix inversion.
Zeroth-order methods favor flat minima in machine learning.
problem Finding solutions with small Hessian trace in optimization.
method Zeroth-order optimization with two-point estimator.
result Zeroth-order optimization converges to flat minima.
New algorithm for safer machine learning with different testing and training distributions.
problem Challenges in modern machine learning where training and testing distributions differ.
method First-order optimization algorithm for superquantile-based learning.
result Promising numerical results show the approach's effectiveness.
Zeroth-order optimization is an important research topic in machine learning. In recent years, it has become a key tool in black-box adversarial attack to neural network based image classifiers. However, existing zeroth-order optimization algorithms rarely extract second-order information of the model function. In this…
A new method for optimizing deep neural networks using TKFAC.
problem Optimizing deep neural networks with second-order methods.
method Proposes Trace-restricted Kronecker-factored Approximate Curvature (TKFAC) for Fisher information matrix approximation.
result TKFAC improves performance on deep network architectures compared to state-of-the-art algorithms.
Exact second-order optimization for deep learning reduces computational cost and improves performance.
problem Inadequate use of second-order optimization methods in deep learning due to high computational cost and non-convexity.
method Developed an exact stochastic second-order Newton method that addresses the non-convexity issue and provides an expression for the stochastic Hessian.
result Exact second-order Newton direction formula and its application in deep learning datasets.
This paper reviews zeroth-order optimization in signal processing and machine learning.
problem Optimization problems without gradient information.
method Iterative steps: gradient estimation, descent direction computation, solution update.
result Demonstrates applications in robustness evaluation and black-box model explanations.
New RL method handles large state-action spaces with complex models.
problem Complex models and large state-action spaces in reinforcement learning.
method π-KRVI, an optimistic modification of least-squares value iteration using kernel ridge regression.
result First order-optimal regret guarantees under general settings, improving over state of the art.
Improves zeroth-order optimization for private machine learning with public data.
problem High computation and memory cost of first-order DP methods.
method PAZO (Public Data Assisted Zeroth-order Optimization) framework.
result Achieves superior privacy/utility tradeoffs across tasks.
This paper studies poisoning attacks in episodic RL and discovers their effectiveness depends on reward bounds.
problem Understanding security threats to RL algorithms through poisoning attacks.
method Examined two types of poisoning attacks: reward and action manipulation, in bounded and unbounded reward settings.
result The effectiveness of poisoning attacks depends on reward bounds, with different attack costs and success rates.
Kernel-based bandit is an extensively studied black-box optimization problem, in which the objective function is assumed to live in a known reproducing kernel Hilbert space. While nearly optimal regret bounds (up to logarithmic factors) are established in the noisy setting, surprisingly, less is known about the noise-f…
We consider the problem of optimizing a high-dimensional convex function using stochastic zeroth-order queries. Under sparsity assumptions on the gradients or function values, we present two algorithms: a successive component/feature selection algorithm and a noisy mirror descent algorithm using Lasso gradient estimate…
Certified algorithms optimize functions with varying costs, providing error bounds.
problem Optimizing functions with varying evaluation costs and error bounds.
method Formalized as a min-max game, proposed certified MFDOO algorithm with cost complexity bound.
result Proposed certified MFDOO algorithm has near-optimal cost complexity for Lipschitz functions.
Bayesian optimization algorithm reduces regret with efficient region pruning.
problem Sequential optimization of unknown functions in high-dimensional spaces.
method Gaussian process-based, domain shrinking through tree-based region pruning.
result Order-optimal regret performance with reduced computational complexity.
The paper debiases mini-batch approximations in deep learning for more accurate optimization and uncertainty quantification.
problem Bias in mini-batch approximations distorts the shape of quadratic approximations used in deep learning.
method Developed and evaluated debiasing strategies for mini-batch approximations.
result Debiasing strategies improve the accuracy of second-order optimization and uncertainty quantification in deep learning.
ConMeZO speeds up zeroth-order optimization for large language models.
problem Slow convergence in high-dimensional parameter spaces of large language models.
method Adaptive directional sampling in a cone centered around a momentum estimate.
result Achieves the same convergence rate as MeZO but up to 2X faster.
Random exploration optimizes Bayesian optimization with optimal error rates and computational efficiency.
problem Optimizing Gaussian Process models in Bayesian optimization.
method Random sampling from a distribution in an infinite dimensional Hilbert space, with domain shrinking and order-optimal regret guarantees.
result Achieves optimal error rates and computational efficiency in both noise-free and noisy settings.
Method extracts features from signals for classification with explainability.
problem Lack of interpretability in signal classification models.
method Combining scattering transform and multiclass logistic regression with zeroth-order optimization.
result Uncovered the meaning of scattering transform coefficients.
Efficient Bayesian FL method improves predictive accuracy and uncertainty estimates.
problem Federated Learning with model and predictive uncertainty and personalization.
method Second-order optimization approach for Bayesian FL.
result Improved predictive accuracies and uncertainty estimates.
New optimization method improves generalization across various tasks.
problem Improving zeroth-order optimization for better generalization.
method Exponential tilting objective to connect zeroth-order optimization with sharpness-aware minimization.
result Achieves better generalization compared to vanilla zeroth-order baselines.
SLEDGE algorithm reduces gradient computation errors in optimization.
problem Accumulated errors in gradient estimation methods for large-scale optimization.
method Single-loop method for finite-sum nonconvex optimization without periodic gradient refresh.
result Achieves nearly optimal gradient complexity and second-order optimality.
Zeroth-order optimization methods lack inherent privacy guarantees.
problem Ensuring differential privacy in zeroth-order optimization methods.
method Analyzing ZO-GD with and without random initialization for convex and strongly convex objectives.
result ZO-GD is not differentially private for strongly convex objectives and can have superlinear privacy loss.
Efficient RNN algorithm guarantees convergence in online learning.
problem Online nonlinear regression with RNNs.
method First-order training algorithm with convergence guarantee.
result The algorithm converges to optimum network parameters.
In this paper, we implement the Stochastic Damped LBFGS (SdLBFGS) for stochastic non-convex optimization. We make two important modifications to the original SdLBFGS algorithm. First, by initializing the Hessian at each step using an identity matrix, the algorithm converges better than original algorithm. Second, by pe…
BMM algorithm improves convergence for nonconvex optimization problems.
problem Constrained nonsmooth nonconvex optimization problems.
method Block majorization-minimization with diminishing radius.
result Improved convergence rate for nonconvex optimization problems.
Optimizes hyperparameter tuning for models using approximate leave-one-out cross-validation.
problem Finding optimal hyperparameters for regularized models using approximate leave-one-out cross-validation.
method Derive efficient formulas for gradient and hessian of approximate leave-one-out cross-validation, apply second-order optimization.
result Demonstrates the effectiveness of the approach on real-world data sets.
We study online learning when partial feedback information is provided following every action of the learning process, and the learner incurs switching costs for changing his actions. In this setting, the feedback information system can be represented by a graph, and previous works studied the expected regret of the le…
DAL enhances black-box bandit algorithms for non-stationary environments.
problem Non-stationary environments in bandit problems.
method DAL combines any stationary bandit algorithm with a change detector.
result DAL consistently outperforms state-of-the-art methods in various non-stationary scenarios.
We consider distributed optimization problems where forming the Hessian is computationally challenging and communication is a significant bottleneck. We develop unbiased parameter averaging methods for randomized second order optimization that employ sampling and sketching of the Hessian. Existing works do not take the…
Derives FACT, an alternative to NFA for neural networks, explaining feature learning.
problem Understanding how neural networks learn representations.
method First-principles approach using first-order optimality conditions.
result FACT explains why NFA holds and provides a principled alternative.
SGD's performance improves with critical batch size, minimizing SFO complexity.
problem Optimizing SGD's performance with batch size and learning rate.
method Analysis of SGD using constant and decaying learning rates, focusing on batch size effects.
result SGD with critical batch size minimizes SFO complexity.
Data processing inequalities link Fisher information to local differential privacy constraints.
problem Understanding how Fisher information scales with local differential privacy constraints.
method Developed data processing inequalities for Fisher information under local differential privacy.
result Implications for private estimation with optimal bounds and error rates.
Paper develops fast method for computing optimal transport.
problem Efficient computation of optimal transport distance between distributions.
method Entropy-regularized extragradient method for first-order optimization.
result Achieves state-of-the-art runtime guarantees and good numerical performance.
Muon optimizes training efficiency by improving data retention at large batch sizes.
problem Improving training efficiency and data retention at large batch sizes.
method Introducing Muon, a second-order optimizer, and combining it with muP for efficient hyperparameter transfer.
result Muon outperforms AdamW in retaining data efficiency at large batch sizes, enabling more economical training.
Large-scale distributed training of deep neural networks suffer from the generalization gap caused by the increase in the effective mini-batch size. Previous approaches try to solve this problem by varying the learning rate and batch size over epochs and layers, or some ad hoc modification of the batch normalization. W…
Recently, we proposed to transform the outputs of each hidden neuron in a multi-layer perceptron network to have zero output and zero slope on average, and use separate shortcut connections to model the linear dependencies instead. We continue the work by firstly introducing a third transformation to normalize the scal…
This research proves that quadratic regularized optimal transport can approximate the Laplace-Beltrami operator on smooth manifolds.
problem Approximating the Laplace-Beltrami operator using optimal transport with quadratic regularization.
method Deriving first-order optimal potentials and analyzing the convergence of discrete Laplace operators.
result The discrete Laplace operators converge to the Laplace-Beltrami operator on smooth manifolds.
Paper proposes a second-order method for faster SVI convergence.
problem Poor convergence rate of first-order SVI algorithms.
method Derives Hessian matrix and implements two numerical schemes for efficient second-order SVI.
result Proposed approach achieves faster convergence compared to first-order SVI.
Paper explores challenges in training PINNs and loss landscape effects.
problem Challenges in training Physics-Informed Neural Networks (PINNs) due to loss landscape issues.
method Examined gradient-based optimizers Adam, L-BFGS, and their combination Adam+L-BFGS, and introduced NysNewton-CG (NNCG).
result Adam+L-BFGS outperforms other optimizers, and NysNewton-CG significantly improves PINN performance.
We consider distributed statistical optimization in one-shot setting, where there are m m m machines each observing n n n i.i.d. samples. Based on its observed samples, each machine then sends an O ( log ( m n ) ) O(\log(mn)) O ( log ( mn )) -length message to a server, at which a parameter minimizing an expected loss is to be estimated. We propose an alg…