The study analyzes implicit biases in neural networks using backward error analysis.
problem Analyzing implicit biases in multitask and continual learning settings.
method Backward error analysis to compute implicit training biases, deriving modified losses with three terms.
result The conflict term, measuring gradient alignment, is a new quantity in continual learning.
Large learning rates lead to various implicit biases in nonconvex optimization.
problem Understanding the conditions under which large learning rates yield edge of stability, balancing, and catapult phenomena.
method Developed a global convergence theory for nonconvex functions without globally Lipschitz continuous gradient, focusing on functions with good regularity.
result These implicit biases are more likely to occur in functions with good regularity, and large learning rates favor flatter regions.
New insights into network generalization show learning rate affects both norm and sharpness.
problem Understanding the generalization of overparameterized networks.
method Empirical analysis and theoretical proof of the trade-off between norm and sharpness.
result Learning rate influences both norm and sharpness, neither alone minimizes generalization error.
Gradient flow in ReLU networks biases towards generalization but makes them vulnerable to adversarial attacks.
problem Generalization vs. Adversarial Robustness in ReLU Networks
method Analysis of gradient flow in two-layer ReLU networks with clustered data.
result Gradient flow biases towards generalization but also makes networks vulnerable to adversarial attacks.
Study reveals how neural network biases align with adversarial attack frequencies.
problem Correlation between neural network biases and adversarial attacks.
method Fourier transform analysis of network implicit bias and adversarial perturbations.
result Network bias and adversarial attack frequencies are highly correlated.
Study reveals biases in gradient descent for GLNs, improving neural network performance.
problem Understanding and improving the inductive biases of deep neural networks.
method Derive infinite-time training limit of gated linear networks and generalize to other networks.
result Theoretical framework captures key inductive biases of ReLU networks.
While implicit feedback (e.g., clicks, dwell times, etc.) is an abundant and attractive source of data for learning to rank, it can produce unfair ranking policies for both exogenous and endogenous reasons. Exogenous reasons typically manifest themselves as biases in the training data, which then get reflected in the l…
Gradient matching method estimates implicit regularization in complex deep learning systems.
problem Estimating implicit regularization in modern deep learning systems with complex modifications.
method Gradient matching methods to empirically estimate implicit regularization.
result Empirical estimation of implicit regularization in arbitrary networks, including dropout.
Gradient descent implicitly regularizes neural networks by penalizing large loss gradients.
problem How to optimize deep neural networks without explicit regularization.
method Backward error analysis to calculate implicit gradient regularization and demonstrate its effectiveness empirically.
result Implicit gradient regularization biases gradient descent toward flat minima, improving model robustness and test errors.
Gradient descent implicitly favors group sparsity in neural networks.
problem Understanding implicit regularization in neural networks for structured sparsity.
method Novel neural reparameterization for diagonally grouped linear networks.
result Gradient descent without explicit regularization biases towards group sparsity.
Kernel semi-implicit variational inference improves variational inference without additional optimization.
problem Intractability of hierarchical semi-implicit distributions in variational inference.
method Kernel semi-implicit variational inference (KSIVI) using kernel methods to eliminate lower-level optimization.
result KSIVI reduces variational inference to kernel Stein discrepancy (KSD) optimization, improving expressiveness and tractability.
Gradient descent biases towards stable rank networks for nearly-orthogonal data.
problem Understanding implicit bias in non-smooth neural networks trained by gradient descent.
method Analysis of two-layer ReLU and leaky ReLU networks trained by gradient descent on nearly-orthogonal data.
result Gradient descent biases towards networks with stable rank and uniform margin for nearly-orthogonal data.
SIP corrects model bias in Bayesian ML.
problem Model selection biases predictions in Bayesian ML.
method Sparse Implicit Processes (SIP) for flexible, trainable predictions.
result SIP provides better predictive distributions than initial models.
Regularized training of an autoencoder typically results in hidden unit biases that take on large negative values. We show that negative biases are a natural result of using a hidden layer whose responsibility is to both represent the input data and act as a selection mechanism that ensures sparsity of the representati…
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.
Kernel SIVI improves variational inference by avoiding lower-level optimization.
problem Intractable densities in semi-implicit variational distributions.
method Kernel SIVI-SM uses a minimax formulation and kernel tricks to avoid lower-level optimization.
result Kernel Stein discrepancy (KSD) objective is computable and leads to convergence guarantees.
Improved Fréchet regression tackles noise and multicollinearity.
problem Addressing noise and multicollinearity in multi-label regression.
method Implicit regularization framework for explicit modeling of relationships.
result Effective modeling of complex dependencies without introducing biases.
New benchmark uncovers hidden biases in LLMs that refuse to answer certain queries.
problem Evaluating fairness in LLMs, especially in sensitive applications.
method Silenced Bias Benchmark (SBB) using activation steering to reduce model refusals during QA.
result Exposes hidden unfair preferences in LLMs' latent space, distinguishing direct responses from underlying fairness issues.
This study investigates how gradient-based methods bias neural networks trained on high-dimensional data.
problem The implicit biases of gradient-based optimization algorithms in neural networks trained on high-dimensional data.
method Investigation of gradient flow and gradient descent in two-layer fully-connected neural networks with leaky ReLU activations.
result Gradient flow and gradient descent lead to neural networks with low-rank solutions and linear decision boundaries.
Noise in RNNs promotes flatter minima and more stable dynamics.
problem Understanding and optimizing the training of RNNs with noise.
method Formalizing RNNs as stochastic differential equations and analyzing the effect of noise in the hidden states.
result Noise injection in RNNs leads to flatter minima, more stable dynamics, and improved robustness.
Mirror flow in shallow neural networks shows similar implicit bias to gradient flow, with key differences in curvature penalties.
problem Analyzing implicit bias in shallow neural networks with mirror flow.
method Characterization through variational problems and scaled potentials.
result Mirror flow with scaled potentials induces a rich class of biases not captured by RKHS norms.
Modality-agnostic compression improves across diverse data types.
problem Efficiently compressing data across multiple modalities.
method Functional view of data, Implicit Neural Representation (INR), modality-agnostic latent representations, variational compression.
result Improved performance compared to existing methods, especially for diverse modalities.
When optimizing over-parameterized models, such as deep neural networks, a large set of parameters can achieve zero training error. In such cases, the choice of the optimization algorithm and its respective hyper-parameters introduces biases that will lead to convergence to specific minimizers of the objective. Consequ…
Over the past two decades, the notion of implicit bias has come to serve as an important component in our understanding of discrimination in activities such as hiring, promotion, and school admissions. Research on implicit bias posits that when people evaluate others -- for example, in a hiring context -- their unconsc…
ReLU networks implicitly favor low-rank solutions, but not as strongly as linear networks.
problem Understanding implicit regularization in ReLU networks for rank minimization.
method Analysis of gradient flow on ReLU networks, empirical testing.
result Gradient flow on ReLU networks does not necessarily minimize ranks, unlike in linear networks.
Model predicts increased social unrest during COVID-19 using social media data.
problem Detecting rising conflict potential in societies during pandemics.
method Neural implicit motive pattern recognition from social media texts.
result Significant increase in conflict indicators during the pandemic.
Proposes a new IPW-based ranking metric for two-sided markets.
problem Addressing bias in implicit user feedback in two-sided markets.
method Extends IPW estimator to two-sided markets, addressing position bias.
result Proposed estimator is unbiased for ground-truth ranking metric.
Graph contrastive learning reveals unique inductive biases.
problem Understanding and optimizing graph contrastive learning methods.
method Systematic study of various GCL methods and their properties.
result GCL methods can work without positive or negative samples, and data augmentations have less impact.
A multi-layer deep Gaussian process (DGP) model is a hierarchical composition of GP models with a greater expressive power. Exact DGP inference is intractable, which has motivated the recent development of deterministic and stochastic approximation methods. Unfortunately, the deterministic approximation methods yield a…
Differentiable NAS frameworks grow networks wider and deeper, revealing biases in wiring evolution.
problem Understanding the evolution of neural architecture wiring in differentiable NAS methods.
method Unified view on searching algorithms, local cost minimization, empirical and theoretical analyses.
result Implicit inductive biases cause observed searching patterns in differentiable NAS methods.
Language models perform worse with implicit reward models than explicit ones.
problem Understanding why implicit reward models generalize worse than explicit ones.
method Investigated the root cause of the generalization gap between IM-RMs and EX-RMs.
result Implicit reward models rely more on superficial token-level cues, leading to worse generalization.
The growing capability and accessibility of machine learning has led to its application to many real-world domains and data about people. Despite the benefits algorithmic systems may bring, models can reflect, inject, or exacerbate implicit and explicit societal biases into their outputs, disadvantaging certain demogra…
It remains a puzzle that why deep neural networks (DNNs), with more parameters than samples, often generalize well. An attempt of understanding this puzzle is to discover implicit biases underlying the training process of DNNs, such as the Frequency Principle (F-Principle), i.e., DNNs often fit target functions from lo…
Industrial recommender systems deal with extremely large action spaces -- many millions of items to recommend. Moreover, they need to serve billions of users, who are unique at any point in time, making a complex user state space. Luckily, huge quantities of logged implicit feedback (e.g., user clicks, dwell time) are …
Local convolutions bias neural networks towards high-frequency adversarial examples.
problem High-frequency adversarial examples in neural networks.
method Analysis of different linear and nonlinear architectures, focusing on the impact of local convolution operations.
result Local convolutions induce an implicit bias towards high frequency features, leading to high-frequency adversarial examples.
JKO scheme adds deceleration in rapidly changing metric curvature directions.
problem Understanding the implicit bias of the JKO scheme in Wasserstein gradient flow.
method Characterized the implicit bias of the JKO scheme at second order in η, modifying the energy functional.
result JKO scheme adds deceleration in directions where metric curvature of J is rapidly changing.
Early alignment in neural networks leads to sparse representations but hinders convergence.
problem The implicit bias of gradient descent during early training phases.
method Quantitative description of early alignment phase in small initialisation, one hidden layer networks.
result Early alignment induces a sparse representation but also hinders convergence to global minima.
Adam avoids simplicity bias in neural networks, leading to better generalization.
problem Simplicity bias in neural networks trained with SGD.
method Comparison of Adam and GD on binary classification tasks with Gaussian data.
result Adam leads to richer and more diverse features, improving generalization.
New method compares community detection algorithms without ground truth.
problem Comparing community detection algorithms without ground truth.
method Integrates any objective with its implicit generative model.
result More expressive methods consistently outperform others.
The key idea behind the unsupervised learning of disentangled representations is that real-world data is generated by a few explanatory factors of variation which can be recovered by unsupervised learning algorithms. In this paper, we provide a sober look at recent progress in the field and challenge some common assump…
SGD converges to critical points of normalized margin in late-stage training for homogeneous neural networks.
problem Analyzing the implicit bias of SGD on homogeneous neural networks.
method Interpreting SGD dynamics as an Euler-like discretization of a conservative field flow associated with the normalized classification margin.
result Normalized SGD iterates converge to the set of critical points of the normalized margin at late-stage training.
Gradient descent biases neural networks to use an average of features, leading to non-robustness.
problem Non-robustness in neural networks due to feature averaging.
method Theoretical analysis and experiments on binary classification tasks.
result Gradient descent trains networks to rely on an average of features, making them vulnerable to adversarial attacks.
We propose NetGAN - the first implicit generative model for graphs able to mimic real-world networks. We pose the problem of graph generation as learning the distribution of biased random walks over the input graph. The proposed model is based on a stochastic neural network that generates discrete output samples and is…
Theoretical justification for deep networks' performance with regularization techniques.
problem Understanding the performance of deep networks trained with the square loss.
method Analysis of gradient flow and theoretical justification of regularization techniques.
result Convergence to solutions with smaller Frobenius norms leads to better classification error bounds.
New method stabilizes machine learning for physics-informed inverse problems.
problem Reconstructing physical quantities from PDE-compliant measurements.
method Physics-informed learning with smooth inductive bias.
result PDE operators stabilize variance and prevent overfitting in fixed dimensions.
SPADE-S improves time series forecasting accuracy for low-magnitude and sparse data.
problem Challenges in forecasting time series with strong heterogeneity in magnitude and sparsity.
method SPADE-S is a robust forecasting architecture that reduces biases and improves overall prediction accuracy.
result SPADE-S outperforms existing state-of-the-art approaches across diverse use cases, improving forecast accuracy by up to 15%.
Why deep neural networks (DNNs) capable of overfitting often generalize well in practice is a mystery [#zhang2016understanding]. To find a potential mechanism, we focus on the study of implicit biases underlying the training process of DNNs. In this work, for both real and synthetic datasets, we empirically find that a…
Gradient descent biases linear models in next-token prediction towards data entropy.
problem Optimization bias in next-token prediction models.
method Analysis of gradient descent on linear models with sparse conditional distributions.
result Gradient descent selects parameters that equate token logits differences to log-odds in the data subspace.