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.
SVAG method examines biased updates in variance-reduced stochastic gradient methods.
problem Examining biased updates in variance-reduced stochastic gradient methods.
method Introduces SVAG, a SAG/SAGA-like method with adjustable bias, analyzed in a cocoercive root-finding setting.
result SVAG's step-size requirements for gradients are less restrictive compared to non-gradient cases, highlighting the need for careful analysis.
We present a convergence rate analysis for biased stochastic gradient descent (SGD), where individual gradient updates are corrupted by computation errors. We develop stochastic quadratic constraints to formulate a small linear matrix inequality (LMI) whose feasible points lead to convergence bounds of biased SGD. Base…
Holdout set improves risk score accuracy without biasing predictions.
problem Updating risk scores can lead to biased estimates when directly applied.
method Use a holdout set of non-intervention population to update risk scores.
result Optimal holdout size reduces adverse outcomes to optimal level.
Simple algorithm approximates rare event frequencies.
problem Approximating the frequency of rare events.
method Iterative update of categorical click-distribution, resulting in a random walk on an n-dimensional simplex.
result The random walk corresponds to a biased Bernoulli convolution under certain conditions.
EigenGame improves eigendecomposition by offering unbiased updates for larger datasets.
problem Minibatch bias in EigenGame limits convergence and parallelism.
method Proposed unbiased stochastic update for EigenGame.
result Asymptotic equivalence to EigenGame, greater parallelism, and improved performance.
Paper analyzes biased stochastic approximation with a novel multistep Lyapunov function.
problem Finite-time analysis of biased stochastic approximation algorithms.
method Developed a multistep Lyapunov function to analyze convergence and error bounds.
result First finite-time error bounds for TD- and Q-learning with linear function approximation.
Bias correction improves language model training performance.
problem Stochastic update bias in preconditioned optimizers.
method Cross-fitted preconditioning and variance-corrected inversion.
result Reduces held-out pretraining loss by 0.15 nats.
Paper proves SHB convergence with biased gradients and approximate step sizes.
problem Establishing convergence of SHB with biased gradients and approximate step sizes.
method Generalizes SHB convergence conditions for biased gradients, approximate step sizes, and block updating.
result Proves convergence of SHB with new conditions for biased gradients and approximate step sizes.
Framework for safely updating machine learning models.
problem Continuous updates to machine learning models can lead to unintended consequences.
method Formalizes the problem as computing the largest locally invariant domain (LID), uses tractable primal-dual formulation.
result Matches or exceeds heuristic baselines for avoiding forgetting while providing formal safety guarantees.
A new method for efficient neural network fine-tuning using queryable low-rank update atoms.
problem Rigidity of static low-rank adaptation methods when input and depth-wise computation vary.
method A shared queryable memory of low-rank update atoms, allowing dynamic and context-sensitive adaptation.
result Improves final test performance and training stability compared to standard low-rank adaptation.
Federated learning improves by unbiased gradient aggregation and controllable meta updating.
problem Gradient biases and inconsistency between target and optimization objectives in federated averaging.
method Unbiased gradient aggregation with keep-trace gradient descent and gradient evaluation strategy, controllable meta updating with small data samples.
result Faster convergence and higher accuracy with different network architectures in various FL settings.
New algorithm detects and discards faulty updates in federated learning.
problem Byzantine failures, biased local datasets, and poisoning attacks in federated learning.
method Adaptive Federated Averaging with Hidden Markov Model for quality update detection.
result Significantly more robust to faulty, noisy, and malicious participants.
New variance-reduction methods solve stochastic composite inclusions.
problem Solving nonmonotone stochastic composite inclusions.
method Developed unbiased and biased variance-reduced estimators for FRBS method.
result Achieved best oracle complexities for finite-sum and expectation settings.
Federated Learning (FL) systems are gaining popularity as a solution to training Machine Learning (ML) models from large-scale user data collected on personal devices (e.g., smartphones) without their raw data leaving the device. At the core of FL is a network of anonymous user devices sharing training information (mod…
Adversarial training adds dynamic perturbations to neural networks for robustness.
problem Accuracy trade-off and lack of diversity in adversarial examples.
method Dynamic adversarial perturbations in the parameter space of neural networks, updating perturbation biases during training.
result Adversarial training with negligible cost and reduced accuracy trade-off.
New framework reduces strategic manipulation cost for minority groups in fair classification.
problem Strategic manipulation disparities in fair classification.
method Constrained optimization framework that constructs classifiers to reduce strategic manipulation cost for minority groups.
result Empirically, the approach reduces strategic manipulation cost for minority groups over multiple real-world datasets.
Analyzes a non-asymptotic SA scheme for non-convex, smooth objectives.
problem Analyzes SA schemes under relaxed assumptions for non-convex, smooth objectives.
method General SA scheme with state-dependent drift and mean field not necessarily gradient type.
result Analyzes the online EM algorithm and policy-gradient method for reinforcement learning.
COMMOD debiases models with minimal and interpretable changes.
problem Inconsistent and costly model updates in fair machine learning.
method Introduced COMMOD, a novel algorithm for algorithmic fairness that minimizes changes and makes them interpretable.
result COMMOD achieves comparable performance to state-of-the-art debiasing methods while making minimal and interpretable changes.
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.
Study on dropout in neural networks using percolation theory.
problem Understanding dropout's effect on neural network training.
method Investigates percolation models mimicking dropout in neural networks.
result Dropout can cause a breakdown in neural networks.
New actor-critic method reduces sample complexity for reinforcement learning.
problem Improving sample complexity for actor-critic algorithms in reinforcement learning.
method Integrates Monte Carlo rollouts into policy search steps for better control over bias.
result Established sample complexity for actor-critic algorithms with policy gradient.
Meta-learning approaches have been proposed to tackle the few-shot learning problem.Typically, a meta-learner is trained on a variety of tasks in the hopes of being generalizable to new tasks. However, the generalizability on new tasks of a meta-learner could be fragile when it is over-trained on existing tasks during …
New method improves matrix factorization speed and accuracy.
problem Matrix factorization optimization problems suffer from biased solutions and lack of convergence guarantees.
method Proposes a novel Bregman distance for matrix factorization, enabling non-alternating schemes with convergence proof.
result Convergence to a stationary point proved for matrix factorization problems.
New method for decentralized learning reduces data and computation needs.
problem High complexity and data/compute requirements for neural network training.
method Asynchronous updates over unreliable network using Distributed Averaging Consensus.
result Models can be learned on highly biased datasets with intermittent communication.
Paper tackles dueling bandits with delayed feedback, revealing preference bias.
problem Real-world dueling bandit applications often face delays in feedback.
method Introduces biased dueling bandit problem with stochastic delayed feedback, presents two algorithms.
result Two algorithms achieve optimal regret bounds for dueling bandit problems with delay.
Paper explores how knowledge distillation transfers inductive biases between models.
problem Transferring inductive biases between models for tasks with limited data.
method Knowledge distillation applied to models with different inductive biases (LSTMs vs. Transformers, CNNs vs. MLPs).
result Effect of inductive biases is transferred through knowledge distillation, impacting both performance and solution characteristics.
Paper proposes SRA algorithm for online learning robustness and adaptivity.
problem Quantifying and evaluating tradeoff between robustness and adaptivity in online learning.
method SRA algorithm using biased stochastic approximation scheme with adaptive threshold.
result SRA algorithm provides superior performance in synthetic and real datasets.
New method discourages models from using bias shortcuts for better generalization.
problem Training models on biased data can lead to poor generalization when bias shifts.
method Train a de-biased representation by encouraging it to differ from biased representations.
result Improved generalization across synthetic and real-world biases.
The paper tackles bandit problems with biased offline data by using causal methods.
problem Improving bandit algorithms with biased offline data that includes confounding and selection biases.
method Formalizes the problem from a causal perspective, categorizes biases, and derives robust bounds for each arm.
result Causal bounds can guide the bandit agent to learn a nearly-optimal decision policy and consistently reduce asymptotic regret.
Machine learning biases in job recommendations can lead to unfair outcomes.
problem Biased recommendations from recommender systems in job matching.
method Addressing biases at various stages of recommender systems training and deployment.
result Techniques can reduce bias in job recommendations, ensuring fair outcomes.
Gradient-based methods can be biased by distributional asymmetries in bivariate categorical data.
problem Gradient-based causal discovery methods can be biased by distributional asymmetries in bivariate categorical data.
method Identified and examined two distributional biases: Marginal Distribution Asymmetry and Marginal Distribution Shift Asymmetry. Employed two simple models to demonstrate and control these biases.
result Gradient-based methods can be biased by distributional asymmetries, and these biases can be controlled.
Study shows statistical biases can mislead transformer models, impairing their generalization.
problem Statistical biases in transformers affect their ability to generalize.
method Evaluated transformer models on synthetic algorithmic tasks with varying statistical biases.
result Statistical biases lead to overestimation of transformer models' generalization capabilities.
New oracles improve stochastic optimization with noisy or biased measurements.
problem Optimizing functions with noisy or biased measurements.
method Introduced biased gradient oracles for stochastic optimization, analyzed RSG and SGD algorithms with these oracles.
result Derived non-asymptotic bounds for convergence rates of algorithms with biased gradient oracles.
Study identifies and measures biases in legal case data.
problem Addressing representation biases and sentencing disparities in legal case data.
method Utilizes two regression models: a baseline and a fair judge model.
result Quantifies biases across demographic groups in criminal data from Cook County (Illinois).
Optimizes explanations for linear models to enhance trust and interpretability.
problem Lack of well-defined interpretability for linear models and ad hoc approaches.
method Optimization framework using coordinate updates on coefficients.
result Derives a family of interpretability metrics for linear models.
Study reveals biases in ImageNet models are not sufficient for generalization.
problem Understanding and improving generalization of image classification models.
method Large-scale study of 48 ImageNet models trained via various methods.
result Biases identified in ImageNet models do not fully explain generalization.
Study examines whether biased training data lead to biased algorithms.
problem Tackles the issue of bias in algorithmic decision-making.
method Examines a setting where biased human decision-makers affect training data and labels.
result Clarifies conditions for bias reversal in algorithmic decision rules.
New method improves model robustness to biased data.
problem Learning unbiased models from biased datasets.
method Developed epsilon-SupInfoNCE and FairKL losses.
result Improved performance on biased datasets.
New method learns collective variables using autoencoders for molecular simulations.
problem Learning low-dimensional slow degrees of freedom (collective variables) for molecular simulations.
method Iterative method involving CV learning with autoencoders and reweighting scheme.
result Achieves convergence of learned collective variables.
The monotonic linear interpolation in deep networks often leads to plateaus, revealing biases in optimization.
problem Plateaus in the optimization landscape of deep networks during monotonic linear interpolation.
method Investigated monotonic linear interpolation on deep neural networks, focusing on biases in weights and biases.
result Interpolating weights and biases differently can lead to significant differences in loss and accuracy, revealing biases in optimization.
This paper examines inductive biases in deep reinforcement learning and their impact on performance.
problem The trade-off between inductive biases and performance in deep reinforcement learning.
method Investigated domain-specific components and adaptive solutions in deep reinforcement learning agents.
result Adaptive components can sometimes outperform domain-specific components, but not always.
Fairness constraints can improve accuracy from biased data.
problem Learning from biased training data can produce biased and suboptimal classifiers.
method Examined fairness-constrained ERM and other recovery methods.
result Equal Opportunity fairness constraint combined with ERM provably recovers Bayes Optimal Classifier under various bias models.
IMeL turns RL into SL by interpolating improved experiences.
problem Improving reinforcement learning efficiency and scalability.
method IMeL uses a reservoir of experiences and a NN regressor for interpolation.
result IMeL achieves preliminary results and proposes itself as a baseline.
PCI combines perception and control using Bayesian inference with object-based representations.
problem Separate perception and control in reinforcement learning.
method Joint Perception and Control as Inference (PCI) framework with Object-based Perception Control (OPC).
result OPC achieves good perceptual grouping quality and outperforms baselines in accumulated rewards.
Entity resolution (ER) presents unique challenges for evaluation methodology. While crowdsourcing platforms acquire ground truth, sound approaches to sampling must drive labelling efforts. In ER, extreme class imbalance between matching and non-matching records can lead to enormous labelling requirements when seeking s…
Rescaled ASGD optimizes distributed learning under heterogeneous data.
problem Vanilla ASGD biases towards a frequency-weighted average of local objectives.
method Rescale worker stepsizes by their computation times.
result Rescaled ASGD converges to the correct global objective in fixed-computation model.
The paper investigates how ambiguous data and cognitive biases affect machine learning in humanitarian decision making.
problem Ambiguous data and cognitive biases impact the interpretability of machine learning models in humanitarian decision making.
method The study will explore the effects of data ambiguity and cognitive biases on machine learning algorithms in humanitarian contexts.
result The research aims to uncover the specific ways in which ambiguous data and cognitive biases influence the interpretability of machine learning models in humanitarian decision making.