This paper studies Fenchel-Young losses, a generic way to construct convex loss functions from a regularization function. We analyze their properties in depth, showing that they unify many well-known loss functions and allow to create useful new ones easily. Fenchel-Young losses constructed from a generalized entropy, …
Cross-entropy loss linked to metric learning, outperforming complex pairwise losses.
problem Improving metric learning performance without complex optimization schemes.
method Theoretical analysis linking cross-entropy to pairwise losses, showing cross-entropy as an upper bound and equivalent to mutual information maximization.
result Minimizing cross-entropy is equivalent to maximizing mutual information, leading to state-of-the-art performance.
Theoretical analysis of cross-entropy loss functions and their robustness.
problem Guarantees for using cross-entropy as a surrogate loss function.
method Theoretical analysis of a broad family of loss functions, including cross-entropy.
result First H-consistency bounds for comp-sum losses and smooth adversarial comp-sum losses. Structured entropy improves classification performance on structured targets.
problem Cross-entropy loss fails to account for target variable structure.
method Proposes structured entropy, a generalization of entropy using random partitions.
result Structured cross-entropy loss yields better results on classification problems with known structure.
We propose a max-pooling based loss function for training Long Short-Term Memory (LSTM) networks for small-footprint keyword spotting (KWS), with low CPU, memory, and latency requirements. The max-pooling loss training can be further guided by initializing with a cross-entropy loss trained network. A posterior smoothin…
Study of focal-entropy for class-imbalanced classification.
problem Understanding the focal-loss in class-imbalanced settings.
method Distributional viewpoint and information-theoretic analysis of focal-entropy.
result Focal-entropy minimizer exists and departs from data distribution.
Proposes squentropy loss for improved classification accuracy and model calibration.
problem Theoretical and empirical evidence for cross-entropy loss is lacking.
method Introduces squentropy loss as the sum of cross-entropy and average square loss over incorrect classes.
result Squentropy loss outperforms cross-entropy and rescaled square losses in classification accuracy and model calibration.
Square loss performs comparably or better than cross-entropy in neural architectures for various tasks.
problem The superiority of cross-entropy loss over square loss in classification tasks is debated.
method Comparison of several neural architectures on NLP, ASR, and computer vision datasets using both loss functions.
result Square loss often produces better results in the majority of tasks, especially in NLP and ASR.
Quantum machine learning uses quantum cross entropy to minimize loss, but measurement loss affects this process.
problem Quantum machine learning's loss minimization through cross entropy is affected by measurement outcomes.
method Defined quantum cross entropy, proved its lower bounds, and investigated its relation to quantum fidelity and likelihood.
result Quantum cross entropy is lower-bounded by negative log-likelihood when derived from quantum data, but measurement outcomes can cause loss.
We present the Tamed Cross Entropy (TCE) loss function, a robust derivative of the standard Cross Entropy (CE) loss used in deep learning for classification tasks. However, unlike other robust losses, the TCE loss is designed to exhibit the same training properties than the CE loss in noiseless scenarios. Therefore, th…
In deep neural network, the cross-entropy loss function is commonly used for classification. Minimizing cross-entropy is equivalent to maximizing likelihood under assumptions of uniform feature and class distributions. It belongs to generative training criteria which does not directly discriminate correct class from co…
PAC-Bayesian bounds for MLPs with cross entropy loss validated.
problem Generalization bounds for MLPs with cross entropy loss.
method Introduced probabilistic explanations and proved PAC-Bayesian bounds using ELBO.
result MLPs with cross entropy loss inherently guarantee PAC-Bayesian generalization bounds.
Convolution Neural Networks (CNN) have recently achieved state-of-the art performance on handwritten Chinese character recognition (HCCR). However, most of CNN models employ the SoftMax activation function and minimize cross entropy loss, which may cause loss of inter-class information. To cope with this problem, we pr…
Paper proposes a new loss function for conditional models using soft targets.
problem Improving generalization performance of deep neural networks on supervised classification tasks.
method Introduces a new loss function compatible with soft targets, based on noise contrastive estimation.
result Soft target InfoNCE loss performs on par with cross-entropy baselines and outperforms other losses.
DE-QT detects optimal Q-learning stopping points.
problem Information loss in Q-learning during prolonged training.
method Introducing DE-QT to detect entropy changes in Q-tables.
result DE-QT identifies the best stopping point for Q-learning.
Unified interpretation of softmax cross-entropy and negative sampling for knowledge graph embedding.
problem Lack of theoretical relationship between softmax cross-entropy and negative sampling loss functions in knowledge graph embedding.
method Used Bregman divergence to provide a unified interpretation of the two loss functions.
result Theoretical findings for fair comparison of softmax cross-entropy and negative sampling are derived.
Deep learning has been shown to achieve impressive results in several domains like computer vision and natural language processing. A key element of this success has been the development of new loss functions, like the popular cross-entropy loss, which has been shown to provide faster convergence and to reduce the vani…
Paper proposes AXE loss for non-autoregressive machine translation, improving performance.
problem Challenges in training non-autoregressive models due to lack of autoregressive factors and cross entropy loss penalties.
method Proposes aligned cross entropy (AXE) loss function using a differentiable dynamic program for better word order alignment.
result AXE-based training improves performance on major WMT benchmarks and sets a new state of the art for non-autoregressive models.
Proposes a method to solve deep neural networks' local minimum problem.
problem Local minimum problem in deep neural networks training.
method Transforms cross-entropy loss into risk-averse error criterion, adjusts RSI, and uses convexity region.
result Trained deep learning machine is expected to be inside a global minimum's attraction basin.
In this report, we review the calculation of entropy-regularised Wasserstein loss introduced by Cuturi and document a practical implementation in PyTorch. Code is available at https://github.com/t-vi/pytorch-tvmisc/blob/master/wasserstein-distance/Pytorch_Wasserstein.ipynb
Two things seem to be indisputable in the contemporary deep learning discourse: 1. The categorical cross-entropy loss after softmax activation is the method of choice for classification. 2. Training a CNN classifier from scratch on small datasets does not work well. In contrast to this, we show that the cosine loss fun…
The paper sets lower bounds for adversarial robustness in multiclass classification.
problem Adversarial robustness in multiclass classification with arbitrary loss functions.
method Dual and barycentric reformulations for robust risk minimization.
result Sharp lower bounds for adversarial risks are computed efficiently.
Proposes SOVR loss to improve adversarial robustness by increasing logit margins.
problem Adversarial training's difficulty in robustness against sophisticated attacks.
method Introduces SOVR loss function that switches from cross-entropy to one-vs-the-rest loss for important samples.
result SOVR loss increases logit margins of important samples, improving robustness against Auto-Attack.
A new loss function improves neural networks' out-of-distribution detection without side effects.
problem Neural networks struggle with out-of-distribution detection due to SoftMax loss issues.
method Proposes IsoMax loss replacing SoftMax loss, maintaining high entropy and fast inferences.
result Significantly improves neural networks' out-of-distribution detection performance.
Proposes MGCE for improved classification performance.
problem Optimizing between robustness and optimization difficulty in classification.
method Minimax formulation of GCE leading to convex optimization over margins.
result MGCE achieves strong accuracy and better calibration, especially in noisy labels.
Gradient descent converges with arbitrary stepsize for separable data under Fenchel-Young losses.
problem Understanding the conditions under which gradient descent converges with arbitrary stepsize.
method Using Fenchel-Young losses and leveraging the classical perceptron argument to derive convergence rates.
result GD converges with arbitrary stepsize for a majority of Fenchel-Young losses, with better rates for specific loss functions.
The aim of this paper is to provide new theoretical and computational understanding on two loss regularizations employed in deep learning, known as local entropy and heat regularization. For both regularized losses we introduce variational characterizations that naturally suggest a two-step scheme for their optimizatio…
We identify a class of over-parameterized deep neural networks with standard activation functions and cross-entropy loss which provably have no bad local valley, in the sense that from any point in parameter space there exists a continuous path on which the cross-entropy loss is non-increasing and gets arbitrarily clos…
In this paper, we focus on the separability of classes with the cross-entropy loss function for classification problems by theoretically analyzing the intra-class distance and inter-class distance (i.e. the distance between any two points belonging to the same class and different classes, respectively) in the feature s…
Given a task of predicting Y from X, a loss function L, and a set of probability distributions Γ on (X,Y), what is the optimal decision rule minimizing the worst-case expected loss over Γ? In this paper, we address this question by introducing a generalization of the principle of maximum entropy. Applying t…
New risk measure considers horizon risk and interest rate uncertainty.
problem Dynamic risk evaluation considering horizon risk and interest rate uncertainty.
method Introduced a risk measure based on generalized Tsallis entropy.
result New q-entropic risk measure quantifies capital requirement.
Given two networks with the same training loss on a dataset, when would they have drastically different test losses and errors? Better understanding of this question of generalization may improve practical applications of deep networks. In this paper we show that with cross-entropy loss it is surprisingly simple to ind…
This paper generalizes BO uncertainty measures using decision-theoretic entropies.
problem Efficiently inferring optima of expensive black-box functions.
method Introduces a generalized entropy measure from statistical decision theory to optimize Bayesian optimization.
result Demonstrates strong empirical performance across various sequential decision-making tasks.
Improved machine learning method estimates entropy production robustly.
problem Estimating entropy production from trajectory data.
method Variational representation of α-divergence loss functions. result Optimal α=−0.5 yields best performance. Consider a financial market in which an agent trades with utility-induced restrictions on wealth. For a utility function which satisfies the condition of reasonable asymptotic elasticity at −∞ we prove that the utility-based super-replication price of an unbounded (but sufficiently integrable) contingent claim i…
Generative Cross-Entropy improves classification with fewer labels.
problem Limited sample efficiency of cross-entropy loss in data-scarce scenarios.
method Proposes Generative Cross-Entropy (GenCE), a new loss function that incorporates generative principles into a standard discriminative network.
result Generative Cross-Entropy outperforms traditional cross-entropy loss across various datasets and conditions.
The paper studies entropy calibration in language models and finds that miscalibration improves slowly with scale.
problem The problem is whether language model entropy calibration improves with scale and if it's possible to calibrate without reducing log loss.
method The authors study a simplified theoretical setting to characterize miscalibration scaling behavior and measure it empirically in language models ranging from 0.5B to 70B parameters.
result The observed scaling behavior of miscalibration is similar to theoretical predictions, indicating slow improvement with scale. The authors also prove theoretically that it is possible to reduce entropy while preserving log loss if access to a black box predicting future entropy is available.
Study minimax regret in sequential probability assignment with and without side information.
problem Minimax regret analysis in sequential probability assignment.
method Upper and lower bounds on minimax regret using square-root entropy.
result Lower bound matches upper bound for Donsker classes, up to log factors.
Improves policy optimization with polylog(T) regret bounds for stochastic losses.
problem Improves theoretical guarantees for policy optimization in stochastic settings.
method Leverages Tsallis and Shannon entropy regularizers for polylog(T) regret, and log-barrier regularizer for adversarial settings.
result Achieves a first-order polylog(T) regret bound for policy optimization in stochastic settings.
This work improves adversarial robustness by boosting model ensembles with margin maximization.
problem Single models are insufficient for defending against adversarial attacks.
method Margin-boosting approach to learn ensembles with maximum margin.
result Our algorithm outperforms existing ensembling techniques and large models trained end-to-end.
New method learns stochastic thermodynamics from system currents.
problem Understanding entropy production in complex dynamical systems.
method Constructs learning framework using currents and machine learning loss functions.
result Derives loss functions for thermodynamic functions directly from dynamics.
Improved loss functions adapt to weight-space anisotropy, outperforming isotropic counterparts.
problem Adapting to the anisotropic nature of deep weight spaces for better performance.
method Refined local entropic loss functions restricted to a subset of weights, exploiting anisotropy.
result Partial local entropies outperform isotropic counterparts on image classification tasks.
Unified approach to structured prediction combining entropy regularization and neuro-symbolic logic.
problem Structured prediction challenges due to large output spaces and insufficient labeled data.
method Neuro-symbolic entropy regularization loss that restricts entropy regularization to valid structures.
result Models predict more accurately and are more likely to be valid.
New bounds on minimax regret for sequential probability assignment using logarithmic loss.
problem Minimizing regret in sequential probability assignment against arbitrary experts.
method Using self-concordance property of logarithmic loss to derive tight bounds.
result Tight bounds on minimax regret for various expert classes.
This work shows that supervised contrastive learning achieves similar results to cross-entropy but requires more iterations.
problem The question of whether there are fundamental differences in representation geometry between supervised contrastive learning and cross-entropy.
method The authors prove that both losses attain their minimum when representations of each class collapse to the vertices of a regular simplex, and they empirically validate this finding.
result Supervised contrastive learning requires more iterations to reach a close-to-optimal state compared to cross-entropy, indicating different optimization behavior.
The paper explores how different loss functions impact reinforcement learning algorithms.
problem Improving reinforcement learning algorithms by optimizing loss functions.
method Comprehensive survey on loss functions in reinforcement learning, proving the benefits of specific loss functions.
result Binary cross-entropy loss leads to first-order bounds and is more efficient than squared loss.
Paper develops a robust HVA measure for dynamic hedging under liquidity stress.
problem Valuation of dynamic hedging under liquidity stress.
method Defines robust HVA as worst-case expected loss over a relative-entropy neighborhood of loss distributions for no-trade bands.
result Wider no-trade bands lower rebalancing costs but increase hedge-error risk.
This work includes a number of novel contributions for the multiple-source adaptation problem. We present new normalized solutions with strong theoretical guarantees for the cross-entropy loss and other similar losses. We also provide new guarantees that hold in the case where the conditional probabilities for the sour…