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.
New neural network approach using mutual information.
problem Training neural networks for imbalanced datasets.
method Converts neural network classifiers to mutual information evaluators.
result New form of softmax leads to better classification accuracy, especially for imbalanced datasets.
Softmax cross-entropy optimizes mutual information in neural networks.
problem Understanding the relationship between mutual information and classification neural networks.
method Demonstrated that optimizing softmax cross-entropy maximizes mutual information between inputs and labels.
result Softmax cross-entropy can approximate mutual information and highlight relevant image regions.
Study shows MSE with sigmoid can match SCE in classification tasks, especially with noisy data.
problem Inconsistent errors in neural network classification tasks.
method Introduced Output Reset algorithm to use MSE with sigmoid activation.
result MSE with sigmoid activation achieves comparable accuracy and convergence rates to Softmax Cross-Entropy, especially in noisy data scenarios.
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…
A new method speeds up SoftMax normalization for embedding learning.
problem Efficiently learning distributed representations with SoftMax normalization.
method Proposes a linear-time heuristic approximation for mSoftMax(XYT), optimizing cross entropy. result Achieves higher or comparable accuracy to existing methods with lower computational time.
New insights into CE dynamics reveal how Hadamard initialization simplifies softmax.
problem Understanding the dynamics of cross-entropy training loss in deep learning.
method Analyzing a two-layer linear neural network with standard-basis vectors as inputs.
result Gradient flow on cross-entropy converges to neural collapse geometry, proving global convergence.
Recently, fully-connected and convolutional neural networks have been trained to achieve state-of-the-art performance on a wide variety of tasks such as speech recognition, image classification, natural language processing, and bioinformatics. For classification tasks, most of these "deep learning" models employ the so…
Cross-entropy loss together with softmax is arguably one of the most common used supervision components in convolutional neural networks (CNNs). Despite its simplicity, popularity and excellent performance, the component does not explicitly encourage discriminative learning of features. In this paper, we propose a gene…
This research improves neural network uncertainty estimates and reliability.
problem Lack of inherent uncertainty estimates and variability in softmax scores.
method Ensemble-based Dirichlet modeling with method of moments estimator.
result Improved stability and predictive uncertainty estimates.
New ACE cost function encourages diversity in neural networks.
problem Training multiple classifiers with controlled diversity.
method Mathematical derivation and gradient control.
result ACE yields better ensemble results than vanilla.
The computational cost of training with softmax cross entropy loss grows linearly with the number of classes. For the settings where a large number of classes are involved, a common method to speed up training is to sample a subset of classes and utilize an estimate of the loss gradient based on these classes, known as…
LLMs learn peaked distributions slowly due to power-law losses.
problem Slow convergence of loss in training large language models.
method Systematic analysis of toy models and empirical evaluation of LLMs.
result Power-law time scaling with an exponent of 1/3 for learning peaked distributions.
Bayesian approach improves neural network classification accuracy and uncertainty.
problem Overconfidence and lack of uncertainty in softmax for classification tasks.
method Model categorical probability using a random variable with a prior distribution.
result Consistent gains in generalization performance across multiple tasks.
Paper tackles long-tailed labels in classification problems.
problem Imbalanced or long-tailed label distribution in real-world classification problems.
method Logit adjustment applied post-hoc or during training to encourage a large relative margin between rare and dominant labels.
result Unified and generalised techniques for coping with long-tailed labels, improving generalisation and performance.
Study Transformer layers under cross-entropy training using mean field control.
problem Understanding the behavior of Transformer layers in cross-entropy training.
method Continuous-depth mean field control analysis, treating depth as time and layer parameters as controls.
result Derivation of a Pontryagin condition for the limiting population problem, involving the softmax residual.
The Softmax function on top of a final linear layer is the de facto method to output probability distributions in neural networks. In many applications such as language models or text generation, this model has to produce distributions over large output vocabularies. Recently, this has been shown to have limited repres…
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…
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…
This work explains how linear representations in large language models arise from training objectives and gradient descent.
problem Understanding the origins of linear representations in large language models.
method A latent variable model to abstract and formalize concept dynamics, combined with analysis of the softmax cross-entropy objective and gradient descent.
result Linear representations emerge when learning from data matching the latent variable model, and this simple structure suffices to yield linear representations.
We empirically investigate the (negative) expected accuracy as an alternative loss function to cross entropy (negative log likelihood) for classification tasks. Coupled with softmax activation, it has small derivatives over most of its domain, and is therefore hard to optimize. A modified, leaky version is evaluated on…
Metric learning aims at learning a distance which is consistent with the semantic meaning of the samples. The problem is generally solved by learning an embedding for each sample such that the embeddings of samples of the same category are compact while the embeddings of samples of different categories are spread-out i…
Previous work shows that adversarially robust generalization requires larger sample complexity, and the same dataset, e.g., CIFAR-10, which enables good standard accuracy may not suffice to train robust models. Since collecting new training data could be costly, we focus on better utilizing the given data by inducing t…
Gated Recurrent Unit (GRU) is a recently-developed variation of the long short-term memory (LSTM) unit, both of which are types of recurrent neural network (RNN). Through empirical evidence, both models have been proven to be effective in a wide variety of machine learning tasks such as natural language processing (Wen…
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. 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.
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.
A common practice in most of deep convolutional neural architectures is to employ fully-connected layers followed by Softmax activation to minimize cross-entropy loss for the sake of classification. Recent studies show that substitution or addition of the Softmax objective to the cost functions of support vector machin…
Develops a mean-field theory for multi-head self-attention under cross-entropy training.
problem Mean-field analysis of multi-head self-attention under cross-entropy training.
method Mean-field theory for a simplified single-layer causal multi-head self-attention model.
result Proves a static finite-head approximation bound for the optimal risk.
Improves uncertainty estimation and OOD detection in neural networks.
problem Accurate uncertainty estimation and OOD detection in neural networks.
method Investigates one-vs-all and distance-based logit representations for probabilities.
result One-vs-all formulations improve calibration without additional complexity.
Simplified plug-in loss approximates EDL for reliable uncertainty estimation.
problem Efficient and reliable uncertainty estimation in real-world sensor-based learning systems.
method Approximate Dirichlet expected objectives with plug-in losses evaluated at the Dirichlet mean.
result Plug-in losses provide comparable predictive accuracy and selective prediction performance to classical EDL, while being simpler to implement.
Chronicals boosts LLM fine-tuning speed by 3.51x over Unsloth.
problem Memory bottleneck in fine-tuning large language models.
method Four optimizations: fused kernels, cross-entropy reduction, LoRA+, and sequence packing.
result 3.51x speedup on A100-40GB with Qwen2.5-0.5B.
Learning with a primary objective, such as softmax cross entropy for classification and sequence generation, has been the norm for training deep neural networks for years. Although being a widely-adopted approach, using cross entropy as the primary objective exploits mostly the information from the ground-truth class f…
Loss assigns examples to classes and superclasses in hierarchical data.
problem Learning from hierarchical classification problems with known class hierarchy.
method Introduces a loss function that considers the hierarchy of classes, aiming for consistent classification across different granularities.
result Improves accuracy and reduces coarse errors in classification compared to cross-entropy loss.
This work compares and evaluates various sampling methods for neural language models.
problem Lack of systematic comparison and myths about sampling methods.
method Monte Carlo sampling, importance sampling, compensated partial summation, noise contrastive estimation.
result All sampling methods can perform equally well if posterior probabilities are corrected.
Focal loss reduces model curvature for better calibration.
problem Improving model confidence in classification problems.
method Geometric interpretation of focal loss to reduce curvature.
result Focal loss reduces the curvature of the loss surface, enhancing model calibration.
Improved similarity search in embeddings using InfoNCE loss.
problem Improving similarity search in embedding models trained by contrastive learning.
method Introduced a new continuity bound for InfoNCE loss via Gâteaux differentiation, preserving the averaging effect of negative samples.
result Demonstrated that the averaging effect of k negative samples in InfoNCE loss carries over to stabilisation of generalisation error as k grows. Optimizing full likelihoods adapts loss scales and shapes for robust modeling.
problem Rigid loss functions limit model adaptability and robustness.
method Optimize full likelihoods with adjustable parameters.
result Adaptive tuning of loss scales and shapes improves model robustness.
Softmax is an output activation function for modeling categorical probability distributions in many applications of deep learning. However, a recent study revealed that softmax can be a bottleneck of representational capacity of neural networks in language modeling (the softmax bottleneck). In this paper, we propose an…
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.
This paper introduces hyperspherical prototype networks, which unify classification and regression with prototypes on hyperspherical output spaces. For classification, a common approach is to define prototypes as the mean output vector over training examples per class. Here, we propose to use hyperspheres as output spa…
In this paper, we propose a Dual Focal Loss (DFL) function, as a replacement for the standard cross entropy (CE) function to achieve a better treatment of the unbalanced classes in a dataset. Our DFL method is an improvement on the recently reported Focal Loss (FL) cross-entropy function, which proposes a scaling metho…
Understanding and evaluating the robustness of neural networks under adversarial settings is a subject of growing interest. Attacks proposed in the literature usually work with models trained to minimize cross-entropy loss and output softmax probabilities. In this work, we present interesting experimental results that …
Revises logistic-softmax likelihood for Bayesian meta-learning in few-shot classification.
problem Inherent uncertainty in logistic-softmax leads to suboptimal performance in meta-learning.
method Redesigns logistic-softmax likelihood with a temperature parameter for better control of prior confidence.
result Achieves well-calibrated uncertainty estimates and comparable/superior performance on benchmark datasets.
DNLL loss improves deep LDA accuracy and consistency.
problem Pathological solutions in unconstrained Deep LDA.
method Introducing Discriminative Negative Log-Likelihood (DNLL) loss.
result Deep LDA trained with DNLL produces clean latent spaces and better calibrated probabilities.
Paper introduces Balanced Meta-Softmax for better long-tailed visual recognition.
problem Long-tailed distribution mismatch between training and testing data.
method Balanced Meta-Softmax, an unbiased extension of Softmax, using a Meta Sampler.
result Balanced Meta-Softmax outperforms state-of-the-art solutions on visual recognition and instance segmentation.
Paper introduces hierarchical softmax for global hierarchical classification tasks.
problem Improving classification accuracy in tasks with class hierarchies.
method Global hierarchical neural networks using hierarchical softmax.
result Hierarchical softmax outperforms regular softmax in multiple datasets.
Typically, Softmax is used in the final layer of a neural network to get a probability distribution for output classes. But the main problem with Softmax is that it is computationally expensive for large scale data sets with large number of possible outputs. To approximate class probability efficiently on such large sc…