Study Q-learning with constant stepsize, proving convergence and bias, and applying extrapolation.
problem Understanding and optimizing Q-learning with constant stepsize.
method Connecting Q-learning to a Markov chain, proving distributional convergence and bias, applying Richardson-Romberg extrapolation.
result Explicit expression for the linear coefficient of the asymptotic bias and improvement of RR extrapolation method.
Mirror Langevin Algorithm converges with zero bias.
problem Achieving convergence with zero bias in discrete-time sampling.
method Discretization of Mirror Langevin Diffusion and mean-square analysis.
result Mirror Langevin Algorithm converges with zero bias.
The paper improves convergence for linear systems using entropic mirror descent with Polyak stepsizes.
problem Convergence analysis for linear systems with unbounded domain.
method Entropic mirror descent with Polyak stepsizes, sublinear and linear convergence results.
result Generalized convergence result for arbitrary convex functions.
Matrix SMD converges to unique solution minimizing Bregman divergence.
problem High-dimensional multi-output classification and matrix completion problems.
method Stochastic Mirror Descent with matrix parameters and matrix mirror functions.
result Matrix SMD converges exponentially to the unique solution minimizing Bregman divergence.
Self-attention optimizers converge to optimal weights, revealing bias patterns.
problem Understanding the implicit bias in self-attention mechanisms.
method Analysis of gradient-based optimization in self-attention layers.
result Adaptive step-size strategies can accelerate convergence to optimal weights.
Gradient descent stagnates in low-precision, but unbiased rounding schemes improve convergence.
problem Stagnation of gradient descent in low-precision computation.
method Proposed unbiased stochastic rounding schemes that trade zero bias for larger probability of preserving small gradients.
result Unbiased rounding methods typically improve convergence rate of gradient descent for convex problems.
Gradient clipping helps private SGD converge despite potential bias.
problem Gradient clipping in private SGD can bias convergence.
method Theoretical analysis and empirical evaluation of gradient clipping effects.
result Gradient clipping can prevent convergence to stationary points and introduces bias.
Paper analyzes \FedAvg's convergence and introduces a new algorithm to reduce bias.
problem Analyzing convergence and bias in Federated Averaging.
method Markov property, first-order bias expansion, Richardson-Romberg extrapolation.
result Bias in \FedAvg can be decomposed into noise and client heterogeneity components.
Study on nonsmooth contractive SA with constant stepsize and Q-learning.
problem Understanding convergence and bias in nonsmooth contractive SA with different noise types.
method Proposed prelimit coupling technique for steady-state convergence and derived asymptotic bias.
result Asymptotic bias of nonsmooth SA is proportional to the square root of the stepsize.
This work studies the implicit bias of mini-batch SGD in classification.
problem Understanding the implicit bias of mini-batch SGD in multi-class classification.
method Characterizes how batch size, momentum, and variance reduction affect convergence and max-margin behavior under different norms.
result Momentum enables small-batch convergence to an approximate max-margin solution, while variance reduction recovers the exact full-batch bias.
BEMA reduces bias in EMA, leading to faster convergence and better performance.
problem Stochasticity in language model fine-tuning destabilizes training.
method Bias-Corrected Exponential Moving Average (BEMA) augmentation of EMA.
result BEMA leads to significantly improved convergence rates and final performance.
Adam optimizes linear classifiers with separable data.
problem Understanding Adam's implicit bias in linear logistic regression.
method Study of Adam's behavior on linearly separable data.
result Adam converges to a linear classifier with maximum ℓ ∞ \ell_\infty ℓ ∞ -margin. New bounds show SGD can match deterministic gradient descent's convergence rate.
problem Optimizing SGD convergence rate under strong convexity and smoothness.
method Computer-aided Lyapunov analysis, focusing on bias-optimal bounds.
result SGD achieves optimal convergence rate in bias terms for a wide range of step-sizes.
Truncated backpropagation through time (TBPTT) is a popular method for learning in recurrent neural networks (RNNs) that saves computation and memory at the cost of bias by truncating backpropagation after a fixed number of lags. In practice, choosing the optimal truncation length is difficult: TBPTT will not converge …
A new algorithm reduces bias and variance in distributionally robust optimization.
problem Distributionally robust optimization with bias and variance issues.
method Prospect, a stochastic gradient-based algorithm that reduces hyperparameter tuning.
result Prospect achieves linear convergence and 2-3x faster convergence on various benchmarks.
Study on bias and extrapolation in LSA with Markovian data, showing bias reduction with Richardson-Romberg extrapolation.
problem Bias in LSA with constant stepsizes and Markovian data.
method Viewing LSA as a Markov chain, proving convergence and bias expansion, and applying Richardson-Romberg extrapolation.
result Bias is proportional to the stepsize up to higher order terms, and Richardson-Romberg extrapolation reduces the bias.
With the purpose of examining biased updates in variance-reduced stochastic gradient methods, we introduce SVAG, a SAG/SAGA-like method with adjustable bias. SVAG is analyzed in a cocoercive root-finding setting, a setting which yields the same results as in the usual smooth convex optimization setting for the ordinary…
Paper analyzes convergence of decentralized algorithms with noise and bias.
problem Finite time convergence analysis of decentralized stochastic approximation schemes.
method Separated iterates into consensual parts and consensus error; bounded consensus error in terms of stationarity.
result Decentralized SA scheme converges at O ( log T / T ) {\cal O}(\log T/ \sqrt{T} ) O ( log T / T ) rate. Study shows how networks converge to minimum norm solutions with regularization.
problem Interpolating between known regions in shallow ReLU networks.
method Investigates empirical risk minimizers and weight decay regularizers.
result Empirical risk minimizers converge to minimum norm interpolants under specific conditions.
Local adaptive methods in FL can accelerate convergence but introduce bias, which is corrected.
problem The effect of using adaptive optimization methods for local updates in federated learning.
method Proposed correction techniques to overcome the bias introduced by local adaptive methods.
result Correction techniques can achieve faster convergence and higher test accuracy than baseline methods.
Paper proposes CounterSample to improve convergence in LTR models.
problem Large variance in IPS weights slows convergence in LTR models.
method Introduces CounterSample algorithm with provably better convergence.
result CounterSample converges faster than standard IPS-weighted methods.
Corrects bias in random sampling matrices for improved ML methods.
problem Inversion bias in random sampling matrices hampers ML applications.
method Corrects inversion bias for various random sampling methods.
result Establishes local convergence rates for sub-sampled Newton methods.
Batch normalization biases linear models towards uniform margins, improving performance in binary classification.
problem Understanding the implicit bias of batch normalization in linear models and neural networks.
method Analyzing gradient descent convergence on linear models and two-layer CNNs with batch normalization.
result Gradient descent with batch normalization in linear models converges to a uniform margin classifier with an exponential convergence rate.
Gradient descent struggles with learning a single neuron with bias.
problem Learning a single neuron with a bias term in the realizable setting with ReLU activation.
method Theoretical study using gradient descent, characterizing critical points, and providing convergence guarantees.
result Gradient descent faces significant challenges in learning a single neuron with bias, unlike the bias-less case.
Proposes EDM algorithm to accelerate model training in distributed networks.
problem Hindered effectiveness of distributed stochastic optimization algorithms due to data heterogeneity and network sparsity.
method Introduces Exact-Diffusion with Momentum (EDM) algorithm, incorporating momentum techniques to mitigate bias and enhance convergence rate.
result EDM algorithm converges sub-linearly to the optimal solution, radius independent of data heterogeneity, for non-convex objective functions.
New minibatching strategy reduces stochastic gradient bias in optimisation.
problem Reducing bias in stochastic gradient descent.
method Symmetric Minibatching Strategy combined with momentum.
result Reduced stochastic gradient bias from O ( h 2 ) \mathcal{O}(h^2) O ( h 2 ) to O ( h 4 ) \mathcal{O}(h^4) O ( h 4 ) . The paper addresses score-mismatched diffusion models and zero-shot conditional samplers.
problem Theoretical guarantees for score-mismatched diffusion models in zero-shot conditional sampling.
method Theoretical analysis of score-mismatched diffusion models and zero-shot conditional samplers.
result Theoretical performance guarantees with explicit dimensional dependencies for score-mismatched diffusion samplers.
New method improves online covariance estimation for SGD.
problem Improving online covariance estimation for SGD.
method Proposes a de-biased covariance estimator that eliminates second-order derivatives.
result Achieves a convergence rate of n ( α − 1 ) / 2 log n n^{(α-1)/2} \sqrt{\log n} n ( α − 1 ) /2 log n , outperforming existing methods. New findings show neural networks can be fooled by adversarial data, but a simple bias fix works.
problem Neural networks can be misled by adversarial data, rendering existing generalization bounds ineffective.
method Extensive theoretical investigation of a specific dataset using infinitely-wide models, focusing on the Neural Tangent Kernel (NTK).
result A sensible choice of output bias completely mitigates the adversarial phenomenon, revealing phase transitions in accuracy.
Adam's bias shifts from full-batch to max-margin of different norms for separable data.
problem Understanding Adam's implicit bias in the incremental batch setting.
method Analyzing incremental Adam on linearly separable data, constructing datasets, and using a proxy algorithm.
result Incremental Adam can converge to different max-margin classifiers depending on the dataset and batching scheme.
New method improves regression estimates, reducing bias.
problem Omitted variable bias in high-dimensional linear regression.
method Post-Double-Autometrics, an alternative to Post-Double-Lasso.
result Post-Double-Autometrics outperforms Post-Double-Lasso.
Frequency bias affects neural network training on non-uniform data.
problem Understanding how frequency bias impacts neural networks trained on non-uniformly distributed data.
method Used the Neural Tangent Kernel (NTK) model to explore the effect of variable density on training dynamics.
result Convergence time for learning a pure harmonic function depends on the local density at a point.
Study iterative regularization for linear models with convex bias, improving robust sparse recovery.
problem Improving robust sparse recovery with iterative regularization for linear models.
method Primal-dual gradient approach, analyzing convergence in presence of noise, combining regularization and optimization.
result Theoretical results show state-of-the-art performances with computational speed-ups.
AMAGOLD improves stochastic gradient MCMC by infrequent Metropolis-Hastings corrections.
problem Bias in stochastic gradient Hamiltonian Monte Carlo (SGHMC).
method AMAGOLD infrequently uses Metropolis-Hastings corrections to remove bias, with a fixed step size schedule.
result AMAGOLD converges to the target distribution with a fixed, rather than a diminishing, step size, and at most a constant factor slower convergence rate.
The paper proves a margin inequality for separating hyperplanes, useful for analyzing algorithmic bias.
problem Analyzing the implicit bias of algorithms in machine learning.
method Proves a nonsmooth Kurdyka-Lojasiewicz inequality for margin function.
result The bias of algorithm iterates converges at least as fast as the square-root of the margin convergence rate.
Gradient flow on ReLU networks converges to a simple model with few regions.
problem Understanding the dynamics of gradient flow in shallow ReLU networks.
method Analysis of gradient flow dynamics on univariate ReLU neural networks.
result Gradient flow converges to a network with at most O(r) linear regions.
GD at EoS edge minimizes logistic loss without monotonic convergence.
problem Understanding GD's implicit bias at the edge of stability.
method Theoretical analysis of logistic regression with constant stepsize GD.
result GD with any constant stepsize minimizes logistic loss over long time scales.
Paper explains spectral bias in neural networks and its relation to neural tangent kernel.
problem Understanding the spectral bias of deep learning models.
method Decomposes neural network training into eigenfunctions of the neural tangent kernel, proving convergence rates.
result Neural networks learn functions with lower complexity faster, explained by eigenvalues of the neural tangent kernel.
SGD benefits from a directional bias in kernel regression models.
problem Improving generalization in kernel regression models.
method Generalized directional bias property of SGD in kernel regression.
result SGD converges along the eigenvector of the largest eigenvalue of the Gram matrix.
Algorithm bounds causal queries under selection bias.
problem Selection bias affects causal analysis.
method Proposes a new algorithm to address both identifiable and unidentifiable causal queries.
result The likelihood of available data is unimodal, enabling bounds on causal queries.
Gradient descent learns ReLU functions with non-zero bias efficiently.
problem Learning ReLU functions with non-zero bias under Gaussian distributions.
method Gradient descent starting from random initialization.
result Gradient descent achieves near-optimal error with high probability.
Study shows exponential error reduction in multiclass classification without bias-variance trade-off.
problem Multiclass classification with margin conditions.
method Analysis of classification error under hard-margin conditions.
result Exponential decrease in classification error without bias-variance trade-off.
Paper analyzes ECE bias and provides bounds for its estimation.
problem Understanding the estimation bias in ECE for machine learning models.
method Information-theoretic approach to analyze bias in uniform mass and uniform width binning strategies.
result Established upper bounds on ECE estimation bias and optimal number of bins.
Analyze SGD with biased gradients, improving convergence rates and accuracy.
problem Analyzing the convergence of SGD with biased gradients.
method Derive convergence results for smooth non-convex functions and quantify the impact of bias magnitude.
result Improved rates under the Polyak-Lojasiewicz condition and insights into how bias magnitude affects accuracy and convergence.
This paper analyzes and improves convergence in federated learning with biased client selection.
problem Analyzing convergence in federated learning with biased client selection.
method First convergence analysis of federated optimization for biased client selection strategies, proposing Power-of-Choice framework.
result Power-of-Choice strategies converge up to 3 times faster and give 10% higher test accuracy than random selection.
Develops a new method to improve performance in multi-objective learning problems.
problem Gradient bias in multi-objective learning leading to degraded performance.
method Stochastic Multi-objective gradient Correction (MoCo) method that guarantees convergence without increasing batch size.
result Demonstrates effectiveness of MoCo method in simulations on multi-task learning.
Paper addresses underestimation bias in double Q-learning, proposing a method to improve learning performance.
problem Underestimation bias in double Q-learning leading to non-optimal fixed points.
method Proposes a simple approach using approximate dynamic programming to bound the target value.
result Significant improvement in learning performance over baseline algorithms in Atari benchmark tasks.
Stochastic Gradient Descent shows directional bias with moderate learning rates, impacting optimization outcomes.
problem Understanding the bias of SGD with moderate learning rates in practical scenarios.
method Analyzing SGD and GD on an overparameterized linear regression problem.
result SGD converges along large eigenvalue directions, GD along small ones, affecting early stopping outcomes.