Improves confidence calibration in neural networks by smoothing labels based on class similarity.
problem Improving confidence calibration in deep neural networks for safety-critical applications.
method Proposes a novel label smoothing technique where label values are based on similarities with the reference class, using different similarity measurements.
result Consistently outperforms state-of-the-art calibration techniques on various datasets and network architectures.
We consider accurately answering smooth queries while preserving differential privacy. A query is said to be K-smooth if it is specified by a function defined on [−1,1]d whose partial derivatives up to order K are all bounded. We develop an ε-differentially private mechanism for the class of K-smooth queries…
The paper explores the identifiability and interpretability of Gaussian process models using different kernel structures.
problem Identifiability and interpretability issues in Gaussian process models.
method The paper examines both single-output and multi-output Gaussian process models using additive and multiplicative mixtures of Matérn kernels.
result The smoothness of a mixture of Matérn kernels is determined by the least smooth component, and none of the mixing weights or parameters are identifiable.
Two strategies for training network classifiers with feature heterogeneity.
problem Training network classifiers with agents having varying feature sizes and unreliable local decisions.
method Promotes global and local smoothing of classifier outputs.
result Output smoothing makes network classifier dynamics more complex, requiring regularization of parameters.
Paper introduces SLS to improve label smoothing regularization.
problem Improve generalization of neural networks by softening labels.
method Structural Label Smoothing (SLS) to mitigate bias in Bayes error rate.
result Empirical validation shows SLS outperforms traditional label smoothing.
Forest-guided smoothing uses random forest outputs for interpretable local smoothers.
problem Creating interpretable local smoothers from complex random forest outputs.
method Uses random forest outputs to define spatially adaptive bandwidth matrices for a linear smoother.
result Improves interpretability and applicability of random forest outputs for various analyses.
RESTA defends LLMs against jailbreaking attacks by adding random noise to embeddings.
problem Vulnerability of LLMs to jailbreaking attacks that generate harmful outputs.
method Adds random noise to embedding vectors and aggregates during token generation.
result RESTA achieves superior robustness versus utility tradeoffs compared to baseline defenses.
Optimistic bounds for multi-output learning using self-bounding Lipschitz condition.
problem Learning vector-valued functions from supervised data.
method Introducing self-bounding Lipschitz condition and proving optimistic bounds using local Rademacher complexity and Srebro's inequality.
result Minimax optimal generalization bounds for multi-output learning, up to logarithmic factors.
In structured output prediction tasks, labeling ground-truth training output is often expensive. However, for many tasks, even when the true output is unknown, we can evaluate predictions using a scalar reward function, which may be easily assembled from human knowledge or non-differentiable pipelines. But searching th…
URNNs are as expressive as general RNNs with ReLU activations.
problem Expressiveness of URNNs compared to general RNNs.
method Input-output equivalence between URNNs and contractive RNNs with ReLU activations.
result URNNs are as expressive as general RNNs with ReLU activations.
Proposes BATer for improved adversarial example detection.
problem Detecting adversarial examples in neural networks.
method Introduces a Bayesian adversarial example detector (BATer) using random components in a Bayesian neural network.
result BATer outperforms state-of-the-art detectors in adversarial example detection.
We consider the problem of learning a structured multi-task regression, where the output consists of multiple responses that are related by a graph and the correlated response variables are dependent on the common inputs in a sparse but synergistic manner. Previous methods such as l1/l2-regularized multi-task regressio…
Interest in multioutput kernel methods is increasing, whether under the guise of multitask learning, multisensor networks or structured output data. From the Gaussian process perspective a multioutput Mercer kernel is a covariance function over correlated output functions. One way of constructing such kernels is based …
We introduce a novel loss function for training deep learning architectures to perform classification. It consists in minimizing the smoothness of label signals on similarity graphs built at the output of the architecture. Equivalently, it can be seen as maximizing the distances between the network function images of t…
Function approximation from input and output data pairs constitutes a fundamental problem in supervised learning. Deep neural networks are currently the most popular method for learning to mimic the input-output relationship of a general nonlinear system, as they have proven to be very effective in approximating comple…
Optimizes CNNs by directing gradients along output channels.
problem Improving generalization error in CNNs.
method Output-channel directed re-weighted L2 or Sobolev metrics.
result Improves generalization error by optimizing gradients.
New method uses machine learning to estimate sensitivity without binning.
problem Estimating sensitivity of high-dimensional data sets without binning.
method Combines machine-learning classification with likelihood-based inference tests using Kernel Density Estimators.
result Significance estimation is not sensitive to non-smooth probability distributions.
Study confirms learning rates for vector-valued spectral algorithms, proving consistency.
problem Theoretical confirmation of learning rates for vector-valued spectral algorithms.
method Rigorous analysis of learning rates for various vector-valued spectral algorithms, including kernel ridge regression and gradient descent.
result Upper and lower bounds on learning rates for vector-valued spectral algorithms, proving minimax optimality in various scenarios.
Cubic spline smoothing improves interpolation between irregularly sampled data.
problem Interpolation discontinuity in recurrent neural networks for irregularly sampled sequences.
method Cubic spline smoothing compensation module trained end-to-end with ODE-RNN.
result Improves interpolation between irregularly sampled data points.
In this paper, we consider efficient differentially private empirical risk minimization from the viewpoint of optimization algorithms. For strongly convex and smooth objectives, we prove that gradient descent with output perturbation not only achieves nearly optimal utility, but also significantly improves the running …
Optimizes sampling from target distributions with applications to online learning.
problem Optimizing the total variation distance between target and sampled distributions.
method Analyzes the sample complexity of approximate rejection sampling and its applications.
result The optimal total variation distance is given by $ ildeΘ( rac{D}{f'(n)})$.
Enhances random forests by smoothing predictions for better performance.
problem Suboptimal performance due to piecewise constant predictions in random forests.
method Kernel-based smoothing mechanism to introduce local regularity.
result Smoothed random forest model consistently improves predictive performance.
Optimizes graph spectral density learning for large networks.
problem Ad-hoc kernel function and bandwidth selection in graph spectral techniques.
method Maximum Entropy approach to learn a smooth graph spectral density.
result Outperforms comparable iterative spectral approaches on synthetic and real graphs.
Improves Gaussian process models for large datasets.
problem Selecting proper covariance functions in Gaussian processes.
method Nonparametric process convolutions and deep GP models.
result Improves performance on benchmarks for GPs, especially for larger datasets.
Spike-and-Slab Deep Learning (SS-DL) is a fully Bayesian alternative to Dropout for improving generalizability of deep ReLU networks. This new type of regularization enables provable recovery of smooth input-output maps with unknown levels of smoothness. Indeed, we show that the posterior distribution concentrates at t…
In presence of sparse noise we propose kernel regression for predicting output vectors which are smooth over a given graph. Sparse noise models the training outputs being corrupted either with missing samples or large perturbations. The presence of sparse noise is handled using appropriate use of ℓ1-norm along-wi…
Deep, wide ConvResNets can approximate functions and their smoothness.
problem Function approximation and smoothness in deep networks.
method Analyzing ConvResNets, proving their ability to approximate functions and their smoothness.
result Large ConvResNets can approximate functions and exhibit sufficient first-order smoothness.
New method improves robustness of large models without sacrificing accuracy.
problem Improving robustness of large pre-trained models without accuracy loss.
method Multi-scale diffusion denoised smoothing, selectively applying smoothing at multiple noise scales.
result Strong certified robustness at high noise levels with accuracy close to non-smoothed classifiers.
Learning to predict multi-label outputs is challenging, but in many problems there is a natural metric on the outputs that can be used to improve predictions. In this paper we develop a loss function for multi-label learning, based on the Wasserstein distance. The Wasserstein distance provides a natural notion of dissi…
New DP algorithm improves privacy and efficiency for convex optimization.
problem Efficient, DP algorithms for convex optimization with strong excess risk bounds.
method Output perturbation for a broad class of tilted loss functions.
result Near optimal DP excess risk and runtime bounds for convex optimization.
Pairwise Label Smoothing improves deep model generalization by reducing overconfidence.
problem Improving deep model generalization through regularization.
method PLS smooths labels for pairs of samples, learning distribution mass during training.
result PLS significantly outperforms LS and baseline models, reducing up to 30% classification error.
Vision transformers benefit from non-smooth components in adaptation.
problem Understanding the role of non-smoothness in vision transformer adaptation.
method Theoretical analysis and extensive experiments on large-scale vision transformers.
result High plasticity of attention modules and feedforward layers leads to better finetuning performance.
FCNv2 robustness tested under noise and random initial conditions.
problem Assessing AI weather forecasting model robustness to input noise.
method Two experiments with varying noise levels and random initial conditions.
result FCNv2 preserves hurricane features under low to moderate noise, but underestimates intensity and persistence.
Paper tackles robust deep learning from weakly dependent data with unbounded loss and input.
problem Tackles robust deep learning from weakly dependent data with unbounded loss and input.
method Establishes non-asymptotic bounds for expected excess risk under strong mixing and ψ-weak dependence assumptions. result Derives a relationship between bounds and r, and shows convergence rate close to i.i.d. results for r=∞. We introduce a variant of Farber's topological complexity, defined for smooth compact orientable Riemannian manifolds, which takes into account only motion planners with the lowest possible "average length" of the output paths. We prove that it never differs from topological complexity by more than 1, thus showing th…
The paper establishes bounds on the smoothness parameter in Gaussian process interpolation.
problem Estimating the smoothness parameter in Gaussian process models.
method Approximation theory in Sobolev spaces and general theorems on parameter estimation.
result Maximum likelihood estimation recovers the true smoothness for certain classes of functions.
Paper tackles over-smoothing in deep GCNs, proposing DropEdge to improve performance.
problem Over-smoothing reduces expressivity in deep GCNs, especially affecting node classification.
method Theoretical analysis of GCN behavior with depth, proposing DropEdge to alleviate over-smoothing.
result DropEdge improves performance on various GCNs, shallow and deep.
Proposes LsrKD and MrKD to improve neural network training performance.
problem Improving neural network training performance, especially on deep networks.
method Extends Label Smoothing Regularization with Self-Knowledge Distillation, introducing LsrKD and MrKD.
result LsrKD and MrKD significantly improve training performance on deep neural networks.
Study analyzes low-energy behavior of Schrödinger operators with Coulomb potentials.
problem Analyzing the limiting resolvent of Schrödinger operators at low energies.
method Using Vasy's second microlocal approach (Lagrangian approach), uniformly analyzing the resolvent from E=0. result Obtained oscillatory asymptotics for the resolvent output at low energy, differing from short-range cases.
Graphs can be fooled by small edge changes, but this work protects them.
problem Adversaries can manipulate graph data to mislead graph classification models.
method We introduce a smoothed graph classification model with a robustness guarantee.
result The smoothed model maintains consistent predictions under small adversarial perturbations.
Recurrent Neural Networks (RNNs) are among the most popular models in sequential data analysis. Yet, in the foundational PAC learning language, what concept class can it learn? Moreover, how can the same recurrent unit simultaneously learn functions from different input tokens to different output tokens, without affect…
Study Weinstein structures on toric divisors' complements.
problem Understanding Weinstein structures on toric divisors' complements.
method Define a partially-centered condition on Delzant polytopes, develop an algorithm for Weinstein handlebody diagrams.
result Explicit Weinstein structures for complements of smoothed toric divisors.
We study the problem of learning high dimensional regression models regularized by a structured-sparsity-inducing penalty that encodes prior structural information on either input or output sides. We consider two widely adopted types of such penalties as our motivating examples: 1) overlapping group lasso penalty, base…
Paper examines test-time smoothing defenses against adversarial examples.
problem Adversarial examples can mislead neural networks, compromising robustness.
method Evaluates various test-time smoothing defenses on ImageNet.
result Non-monotonic relation between attacks and defenses, large variance among samples.
Transformers handle infinite dimensional inputs effectively by feature extraction and dynamic feature selection.
problem Understanding the approximation and estimation ability of Transformers with infinite dimensional inputs.
method Anisotropic smoothness analysis and feature extraction properties of Transformers.
result Transformers avoid the curse of dimensionality and dynamically select important features.
Proposes NRS to find flat minima in deep neural networks.
problem Finding optimal solutions in deep neural networks with overparameterization.
method NRS leverages the concept of flat minima and uses Kullback-Leibler divergence to regularize the neighborhood region in weight space.
result NRS drives optimizers towards flat minima, improving generalization ability across various model architectures.
DiffDenoise preserves fine structures in medical images using conditional diffusion models.
problem Medical image denoising often results in loss of fine structures.
method Conditional diffusion model with stabilized reverse sampling and supervised training.
result DiffDenoise outperforms state-of-the-art methods in medical image denoising.
We propose an inference method to estimate sparse interactions and biases according to Boltzmann machine learning. The basis of this method is L1 regularization, which is often used in compressed sensing, a technique for reconstructing sparse input signals from undersampled outputs. L1 regularization impedes the …