New method reduces variance in stochastic optimization with high confidence.
problem Achieving high-probability guarantees in stochastic optimization with weaker noise assumptions.
method Stochastic proximal point method combining proximal subproblem solver and probability booster.
result Demonstrates convergence with low sample complexity under bounded variance assumptions.
We present a set of log-price integrated variance estimators, equal to the sum of open-high-low-close bridge estimators of spot variances within n subsequent time-step intervals. The main characteristics of some of the introduced estimators is to take into account the information on the occurrence times of the high a…
Learning in models with discrete latent variables is challenging due to high variance gradient estimators. Generally, approaches have relied on control variates to reduce the variance of the REINFORCE estimator. Recent work (Jang et al. 2016, Maddison et al. 2016) has taken a different approach, introducing a continuou…
Optimizes wide low-rank neural networks for reduced parameters and cost.
problem Reducing the number of learnable parameters in wide neural networks.
method Analyzed edge-of-chaos dynamics and derived formulae for optimal weight and bias variances.
result Optimal weight and bias variances for low-rank networks follow from multiplicative scaling.
Paper introduces VDE, a variance-reduced determinant estimator.
problem Estimating determinants with low variance and efficiency.
method Combines variational inference and spherical normalizing flows.
result VDE achieves zero variance in ideal cases, requiring only one sample.
SkMM selects data for finetuning by balancing bias and variance.
problem Balancing bias and variance in high-dimensional finetuning.
method Gradient sketching for bias reduction and moment matching for variance reduction.
result Gradient sketching selects samples efficiently and accurately.
VarGrad reduces variance in ELBO gradient estimation for variational inference.
problem Improving the variance of gradient estimators in variational inference.
method VarGrad uses a new log-variance loss to estimate the ELBO gradient, achieving lower variance than the score function method.
result VarGrad offers a lower variance gradient estimator compared to other methods.
Due to the high variance of policy gradients, on-policy optimization algorithms are plagued with low sample efficiency. In this work, we propose Augment-Reinforce-Merge (ARM) policy gradient estimator as an unbiased low-variance alternative to previous baseline estimators on tasks with binary action space, inspired by …
Random forests reduce bias and variance, especially in low SNR settings.
problem Reducing bias and variance in machine learning models, particularly in low SNR scenarios.
method Empirical study of random forests and bagging ensembles, focusing on the importance of mtry tuning. result Random forests reduce both bias and variance, outperforming bagging ensembles in high SNR settings.
In earlier studies, the estimation of the volatility of a stock using information on the daily opening, closing, high and low prices has been developed; the additional information in the high and low prices can be incorporated to produce unbiased (or near-unbiased) estimators with substantially lower variance than the …
Low-variance gradient estimation is crucial for learning directed graphical models parameterized by neural networks, where the reparameterization trick is widely used for those with continuous variables. While this technique gives low-variance gradient estimates, it has not been directly applicable to discrete variable…
Paper proposes diagnostics for error and variance estimation in randomized matrix computations.
problem Safe use of randomized matrix algorithms in applications.
method Leave-one-out error estimator and jackknife resampling method.
result Provides rapid diagnostics to assess quality of randomized matrix computations.
Improved outlier detection in hierarchical Gaussian Processes using Wasserstein-2 kernels.
problem Outlier detection limitations in stacked Gaussian Processes.
method Proposed a hybrid kernel combining Euclidean and Wasserstein-2 distances, emphasizing variance in Wasserstein-2 computations.
result Improved performance and enhanced out-of-distribution detection on various datasets.
LAZO reduces query complexity and variance in ZO methods.
problem High query complexity and variance in zeroth-order optimization.
method LAZO uses adaptive lazy queries to reduce variance and save queries.
result LAZO achieves lower regret and query complexity compared to existing methods.
Random forest performance depends on SNR and covariate characteristics.
problem Understanding when random forests perform well.
method Systematic analysis of out-of-sample MSE for different SNR scenarios.
result Randomization effectiveness depends on SNR and covariate characteristics.
ES-Single uses ES to estimate gradients in unrolled graphs, reducing variance and improving performance.
problem Estimating gradients in unrolled computation graphs with low variance and stability.
method Evolution strategies (ES) applied to unrolled graphs, with a single perturbation per particle.
result ES-Single reduces variance compared to PES, leading to better performance in various tasks.
Safe-FinRL uses DRL for high-frequency stock trading, reducing bias and variance.
problem Challenges in applying DRL to high-frequency stock trading, especially bias and variance issues.
method Safe-FinRL separates financial time series into near-stationary short environments and uses Trace-SAC with a general retrace operator.
result Safe-FinRL reduces bias and variance significantly in near-stationary financial environments.
W2S FT often outperforms weak teachers due to low intrinsic dimensionality.
problem Understanding why weak-to-strong finetuning outperforms weak models.
method Analyzing W2S in ridgeless regression setting, focusing on variance reduction.
result Weak teacher's variance is inherited by strong student in shared feature subspace, reduced in discrepancy subspace.
We propose a generic framework based on a new stochastic variance-reduced gradient descent algorithm for accelerating nonconvex low-rank matrix recovery. Starting from an appropriate initial estimator, our proposed algorithm performs projected gradient descent based on a novel semi-stochastic gradient specifically desi…
A new measure k-variance captures local distributional shape.
problem Summarizing distributional shape with local information.
method Random bipartite matchings and stochastic approximation.
result Easily approximated k-variance measures capture local distributional properties. Deep learning models can have low bias and variance, contrary to classical theory.
problem Understanding the performance of deep learning models at high complexity.
method Developed a fine-grained bias-variance decomposition for random feature kernel regression, analyzing the effects of sampling, initialization, and labels.
result The variance terms exhibit non-monotonic behavior and can diverge at the interpolation boundary, even in the absence of label noise.
MLaaS (ML-as-a-Service) offerings by cloud computing platforms are becoming increasingly popular. Hosting pre-trained machine learning models in the cloud enables elastic scalability as the demand grows. But providing low latency and reducing the latency variance is a key requirement. Variance is harder to control in a…
A method learns common bias for multiple low-variance tasks without hyper-parameter tuning.
problem Learning common bias for multiple low-variance tasks without manual tuning.
method Two variants of online learning methods (aggressive and lazy) that update bias after each datapoint or at the end of each task.
result Across-tasks regret bound derived for the method, showing faster rates for aggressive variant and standard rates for lazy variant.
We present Matrix Krasulina, an algorithm for online k-PCA, by generalizing the classic Krasulina's method (Krasulina, 1969) from vector to matrix case. We show, both theoretically and empirically, that the algorithm naturally adapts to data low-rankness and converges exponentially fast to the ground-truth principal su…
Reward models need more than just accuracy for effective RLHF.
problem The effectiveness of reward models in RLHF is not fully understood.
method An optimization perspective to evaluate reward models.
result Reward models with low reward variance can lead to a flat optimization landscape, hindering performance.
CPCR mitigates bias in PCR for overparameterized models.
problem Bias in Principal Component Regression (PCR) for overparameterized models.
method Calibrated Principal Component Regression (CPCR) learns a low-variance prior in the PC subspace and calibrates the model in the original feature space.
result CPCR outperforms standard PCR in overparameterized settings, improving prediction across multiple problems.
To backpropagate the gradients through stochastic binary layers, we propose the augment-REINFORCE-merge (ARM) estimator that is unbiased, exhibits low variance, and has low computational complexity. Exploiting variable augmentation, REINFORCE, and reparameterization, the ARM estimator achieves adaptive variance reducti…
PLUMAGE improves large model training efficiency and stability.
problem Accelerator memory and networking constraints during large model training.
method Probabilistic Low rank Unbiased Minimum Variance Gradient Estimator (PLUMAGE) that resolves bias and variance issues.
result PLUMAGE reduces training loss by 28% on average across the GLUE benchmark.
New mechanisms improve differential privacy for scalar queries.
problem Improving differential privacy for scalar, real-valued query functions.
method Mixing multiple Gaussian distributions to satisfy differential privacy.
result Mechanisms yield lower noise amplitudes and variances compared to the analytic Gaussian mechanism.
Improved LLM pre-training performance through better weight and variance control.
problem Improper weight and variance control in LLM pre-training affects downstream task performance.
method Introduced Layer Index Rescaling (LIR) and Target Variance Rescaling (TVR) techniques.
result Substantial improvements in downstream task performance (up to 4.6%) and reduced extreme activation values.
We study the problem of estimating low-rank matrices from linear measurements (a.k.a., matrix sensing) through nonconvex optimization. We propose an efficient stochastic variance reduced gradient descent algorithm to solve a nonconvex optimization problem of matrix sensing. Our algorithm is applicable to both noisy and…
Recently, the \textit{Tensor Nuclear Norm~(TNN)} regularization based on t-SVD has been widely used in various low tubal-rank tensor recovery tasks. However, these models usually require smooth change of data along the third dimension to ensure their low rank structures. In this paper, we propose a new definition of da…
EVA adapts LoRA for faster, more efficient fine-tuning.
problem Fast and efficient fine-tuning of large models for specific tasks.
method EVA uses directions capturing most activation variance for initialization, maximizing gradient signal and reducing parameters.
result EVA achieves faster convergence and higher average scores across tasks, reducing parameters.
The alternating direction method of multipliers (ADMM) is a powerful optimization solver in machine learning. Recently, stochastic ADMM has been integrated with variance reduction methods for stochastic gradient, leading to SAG-ADMM and SDCA-ADMM that have fast convergence rates and low iteration complexities. However,…
Novel estimator reduces diffusion model variance.
problem High variance in score function estimation for diffusion models.
method Uses nearest neighbour samples to estimate the score function.
result Significant decrease in variance, leading to improved model performance.
We propose a scheme for recycling Gaussian random vectors into structured matrices to approximate various kernel functions in sublinear time via random embeddings. Our framework includes the Fastfood construction as a special case, but also extends to Circulant, Toeplitz and Hankel matrices, and the broader family of s…
Gradient-based methods for optimisation of objectives in stochastic settings with unknown or intractable dynamics require estimators of derivatives. We derive an objective that, under automatic differentiation, produces low-variance unbiased estimators of derivatives at any order. Our objective is compatible with arbit…
Probabilistic principal component analysis (PPCA) seeks a low dimensional representation of a data set in the presence of independent spherical Gaussian noise. The maximum likelihood solution for the model is an eigenvalue problem on the sample covariance matrix. In this paper we consider the situation where the data v…
New algorithms reduce regret in both stochastic and deterministic environments.
problem Designing algorithms that perform well in both types of MDPs.
method Proposed new environment norms and algorithms with variance-dependent regret bounds.
result First algorithm with simultaneously optimal bounds for both stochastic and deterministic MDPs.
Investigates upsampling vs. upweighting for balanced training on skewed datasets.
problem Balancing training on heavily imbalanced datasets with scarce data.
method Theoretical and empirical analysis of upsampling and upweighting strategies.
result Upsampling and upweighting diverge under stochastic gradient descent, with upsampling leading to faster convergence but higher overfitting risk.
A hybrid strategy forecasts short-term loads using Warm-start Gradient Tree Boosting.
problem Lack of effective short-term load forecasting methods.
method Hybrid strategy integrating four different inference models: tree-based ensemble method Warm-start Gradient Tree Boosting (WGTB).
result Demonstrates effectiveness of hybrid strategy on real datasets.
LoCoV reduces portfolio optimization errors from sample covariance matrices.
problem Large errors in sample covariance matrix for optimal portfolio weights.
method LoCoV (low dimension covariance voting) algorithm to reduce these errors.
result LoCoV outperforms classical methods in portfolio optimization experiments.
Paper improves Gumbel-Softmax estimator variance reduction.
problem Challenges in gradient estimation for models with discrete latent variables.
method Rao-Blackwellization applied to straight-through Gumbel-Softmax estimator.
result Reduces mean squared error and variance of Gumbel-Softmax estimator.
New approach for estimating individual treatment effects in low compliance settings.
problem Estimating individual treatment effects in scenarios with low compliance.
method Proposes a new approach using Structural Causal Model and do-calculus to estimate Individual Prescription Effect (IPE) with asymptotic variance guarantees.
result Consistently improves state-of-the-art in low compliance settings.
New methods solve graph sparsity optimization problems faster.
problem Complex graph sparsity optimization problems in disease outbreak monitoring and social network analysis.
method Stochastic variance-reduced gradient-based methods GraphSVRG-IHT and GraphSCSG-IHT.
result Our methods achieve linear convergence speed.
We solve the paradox of score-based methods by minimizing path variance.
problem Score-based methods are path-dependent, leading to inaccurate and unstable estimators.
method Propose MVP Principle to minimize path variance, derive closed-form expression, and use flexible Kumaraswamy Mixture Model.
result Establishes new state-of-the-art results on challenging benchmarks.
We reduce variance in monetization metrics for ranking experiments.
problem Heavy-tailed monetization metrics lead to unreliable conclusions in A/B experiments.
method Post-stratification combined with CUPED.
result Significant reduction in variance and improved decision stability.
Variance reduction (VR) methods boost the performance of stochastic gradient descent (SGD) by enabling the use of larger, constant stepsizes and preserving linear convergence rates. However, current variance reduced SGD methods require either high memory usage or an exact gradient computation (using the entire dataset)…