New method prevents entropy collapse in Transformer training, leading to more stable and robust models.
problem Training instability in Transformers, especially in attention layers.
method Spectral normalization with a learned scalar to prevent entropy collapse.
result Prevents entropy collapse, leading to more stable training.
Self-attention networks localize when eigenspectrum variance is small.
problem Self-attention mechanisms can lead to rank and entropy collapses, reducing model expressivity and trainability.
method Characterized attention localization using query-key eigenspectrum variance.
result Small eigenspectrum variance prevents both rank and entropy collapses, improving model performance.
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.
Gradient flow in softmax models tends to produce low-entropy outputs.
problem Understanding the training dynamics of softmax-based models.
method Analysis of gradient flow dynamics in the value-softmax model.
result Gradient flow drives optimization towards low-entropy solutions.
Mathematical framework for understanding attention in neural networks.
problem Lack of theoretical understanding of attention in neural networks.
method Proposes a measure-theoretic model of attention and interprets self-attention as a system of self-interacting particles.
result Shows that attention is Lipschitz-continuous under suitable assumptions.
Attention to entropic communication improves message decoding and cooperation.
problem Improper weighting in attention mechanisms hinders optimal communication.
method Combining attention and relative entropy (RE) for optimal message encoding and decoding.
result Proper attention communication emerges via weighted RE, aiding optimal protocols.
New analysis shows how cross-entropy training shapes attention in transformers.
problem Understanding how gradient-based learning creates the required internal geometry in transformers.
method Developed a first-order analysis of cross-entropy training effects on attention scores and values in a transformer attention head.
result Introduced an advantage-based routing law and responsibility-weighted update for attention scores and values, respectively.
A new method for feature fusion in U-Net decoders using difference-based gating.
problem Precise fusion of high-level semantics and low-level details in U-Net decoder reconstruction.
method Proposes two difference-based gating approaches: Feature-difference gating (FDG) and Entropy-difference gating (EDG).
result Both FDG and EDG methods outperform existing attention-based fusion methods, with EDG showing superior performance.
Transformers model contextual relations using probabilistic measures, revealing their expressive power.
problem Lack of clear understanding of Transformer's ability to model contextual relations.
method Introduced a measure-theoretic framework connecting softmax attention and entropy-regularized optimal transport.
result Transformer architectures can approximate arbitrary contextual relations, and the choice of normalization affects how these relations are represented.
Unified thermodynamic approach to Transformer attention dynamics.
problem Understanding the statistical mechanics of Transformer attention.
method Constructing a Lagrangian on the information manifold to analyze attention dynamics.
result Establishes a formal correspondence between scaled dot-product attention and canonical ensemble statistics.
Paper proposes a new hierarchical attention mechanism for multi-scale data.
problem Challenges in applying neural attention to multi-scale, multi-modal data.
method Developed a mathematical framework for multi-modal, multi-scale data and derived optimal neural attention mechanics.
result Proposed hierarchical attention mechanism improves transformer performance in multi-scale, multi-modal settings.
Unified theory explains two failure modes of deep transformers and provides initialisation guidelines.
problem Two failure modes (rank collapse and entropy collapse) of self-attention layers in deep transformers.
method Analytical theory of signal propagation through deep transformers, using the Random Energy Model analogy.
result Simple algorithm to compute trainability diagrams for correct initialisation hyper-parameters.
Attention networks show promise for both vision and language tasks, by emphasizing relationships between constituent elements through weighting functions. Such elements could be regions in an image output by a region proposal network, or words in a sentence, represented by word embedding. Thus far the learning of atten…
Unified framework for critical scaling of inverse temperature in self-attention.
problem Conflicting inverse-temperature laws for long-context self-attention.
method Counting gaps and defining an upper-tail accumulation scale.
result Critical inverse-temperature scale determined by gap-counting function.
The celebrated KAM Theory says that if one makes a small perturbation of a non-degenerate completely integrable system, we still see a huge measure of invariant tori with quasi-periodic dynamics in the perturbed system. These invariant tori are known as KAM tori. What happens outside KAM tori draws a lot of attention. …
Hard visual attention is a promising approach to reduce the computational burden of modern computer vision methodologies. Hard attention mechanisms are typically non-differentiable. They can be trained with reinforcement learning but the high-variance training this entails hinders more widespread application. We show h…
A framework uses free probability to analyze Transformer models.
problem Understanding the dynamics and complexity of Transformer-based language models.
method Formal operator-theoretic analysis using free probability theory.
result Entropy-based generalization bounds derived under freeness assumptions.
Paper proposes a new method to learn distribution kernels via entropy maximization.
problem Challenges in applying kernel methods to distribution regression tasks.
method Proposes a novel objective for unsupervised learning of data-dependent distribution kernels based on entropy maximization.
result Demonstrates the effectiveness of the learned kernel across different modalities.
Model improves information transfer from visual streams.
problem Challenges in unsupervised learning from continuous visual data.
method Inspired by physics, maximizes mutual information through temporal process.
result Focus of attention enhances information transfer from input stream.
Study improves materials discovery for high-entropy alloys using sparse linear models.
problem Inefficient materials discovery due to combinatorial explosion in alloy compositions.
method Sparse mixed linear modeling with anchor-based guidance for feature selection and prediction.
result Developed a method that balances predictive performance and interpretability for materials discovery.
Transformer models align words through attention weights, closely approximating Optimal Transport.
problem Understanding the internal mechanism of transformer models in language processing.
method Empirical evidence and theoretical analysis of attention weights and their relation to Optimal Transport.
result Transformer models can simulate gradient descent on the dual of entropy-regularized OT problem, providing a theoretical foundation for token alignment.
Analyzes how BPE tokenisation affects corpus statistics and model entropy in transformer models.
problem Understanding how natural language properties relate to tokenisation schemes in transformer models.
method Analyzes Shannon entropy of corpora under Zipfian distribution, investigates BPE transformations, trains language models, and uses attention diagnostics.
result Transformer models trained on BPE-tokenised corpora increasingly agree with Zipfian predictions as BPE depth increases, indicating reduced local token dependencies.
Financial time series prediction, especially with machine learning techniques, is an extensive field of study. In recent times, deep learning methods (especially time series analysis) have performed outstandingly for various industrial problems, with better prediction than machine learning methods. Moreover, many resea…
Transformer with self-attention has achieved great success in the area of nature language processing. Recently, there have been a few studies on transformer for end-to-end speech recognition, while its application for hybrid acoustic model is still very limited. In this paper, we revisit the transformer-based hybrid ac…
Transformers mimic Bayesian reasoning in controlled settings, revealing geometric mechanisms.
problem Verifying if transformers perform Bayesian reasoning rigorously in natural data.
method Constructing Bayesian wind tunnels with known posteriors and proving memorization impossibility.
result Transformers achieve 10−3-10−4 bit accuracy in Bayesian posteriors, while MLPs fail. Bitcoin's price direction is better predicted without additional drivers during high volatility.
problem Predicting Bitcoin's price direction using various determinants.
method Continuous local transfer entropy for feature selection and deep learning classification model.
result Bitcoin's price direction can be better predicted without additional drivers during high volatility.
SymCircuit learns PC structure via entropy-regularized RL, improving inference efficiency and accuracy.
problem Greedy algorithms in PC structure learning lead to suboptimal solutions.
method Entropy-regularized reinforcement learning to train a learned generative policy for PC structure inference.
result SymCircuit learns the optimal policy as a tempered Bayesian posterior, improving inference efficiency and accuracy.
Pointwise localization allows more precise localization and accurate interpretability, compared to bounding box, in applications where objects are highly unstructured such as in medical domain. In this work, we focus on weakly supervised localization (WSL) where a model is trained to classify an image and localize regi…
CAT framework improves AI medical screening fairness and reliability.
problem Imbalanced data, varying performance across cohorts, and patient-level inconsistencies in traditional metrics.
method CAT framework introduces patient-level assessment, entropy-based distribution weighting, and cohort-weighted sensitivity and specificity.
result Enhanced predictive reliability, fairness, and interpretability of AI-driven medical screening models.
Sequence-to-sequence models, such as attention-based models in automatic speech recognition (ASR), are typically trained to optimize the cross-entropy criterion which corresponds to improving the log-likelihood of the data. However, system performance is usually measured in terms of word error rate (WER), not log-likel…
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.
Stochastic attention learns to retrieve and generate from memory without training.
problem Learning to retrieve and generate from memory efficiently.
method Langevin dynamics on modern Hopfield energy for stochastic attention.
result Stochastic attention can retrieve and generate from memory without training, matching gold standards.
EPSTE: A geometric token and deep learning approach to estimating transfer entropy in neuroimaging time series
problem Inferring directed interactions between neural systems from EEG and MEG
method Reframing TE estimation as a learnable problem operating on structured symbolic representations
result EPSTE achieves near-perfect recovery of ground-truth directed structure and significantly lower absolute error than the baseline
CutMix training technique improves spatial locality in Vision Transformers.
problem Improving spatial locality in Vision Transformers trained from scratch.
method Comparison of Baseline and Modern training protocols on CIFAR-10, CIFAR-100, and Tiny-ImageNet.
result CutMix training component significantly reduces Mean Attention Distance (MAD) in early layers of Vision Transformers.
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.
The study explains how transformer components enable in-context learning.
problem Understanding how transformer components contribute to in-context learning.
method Analyzed a two-attention-layer transformer model trained on Markov chain data.
result Gradient flow converges to a limiting model with a copier, selector, and classifier mechanism.
Adversarial attacks on deep neural networks (DNNs) have been found for several years. However, the existing adversarial attacks have high success rates only when the information of the victim DNN is well-known or could be estimated by the structure similarity or massive queries. In this paper, we propose to Attack on A…
We adapt Shapley values to explain model uncertainty, connecting it to information theory.
problem Explaining uncertainty in model predictions.
method Adapted Shapley value framework to quantify feature contributions to predictive uncertainty.
result Deep connections between Shapley values and information theory quantities.
Imaging techniques are essential tools for inquiring a number of properties from different materials. Liquid crystals are often investigated via optical and image processing methods. In spite of that, considerably less attention has been paid to the problem of extracting physical properties of liquid crystals directly …
DeepDIVE disentangles input into marginal and conditional distributions for multi-task learning.
problem Challenges in multi-task learning due to conflicting objectives.
method Inspired by probability theory, DeepDIVE uses a variational autoencoder with disentangled features and cross-attention mechanism.
result DeepDIVE disentangles input and improves forecast accuracy compared to baseline models.
This work improves VAEs using MCMC methods for better variational bounds.
problem Improving the expressiveness of variational distributions in VAEs.
method Entropy-based adaptation for MALA/HMC chains to optimize tighter variational bounds.
result Higher held-out log-likelihoods and improved generative metrics.
Distances between probability distributions that take into account the geometry of their sample space,like the Wasserstein or the Maximum Mean Discrepancy (MMD) distances have received a lot of attention in machine learning as they can, for instance, be used to compare probability distributions with disjoint supports. …
The article explores how organisms and machines learn and recognize the world using Bayesian inference and thermodynamics.
problem Understanding how organisms and machines learn and recognize the world.
method Introducing a thermodynamic view of the Bayesian brain hypothesis, using a simple generative model of spiking neural populations.
result The process of Bayesian inference can be quantified using entropy, revealing the perceptual capacity of neural activity.
GISST interprets GNNs by combining attention and sparsity for graph structure and node feature importance.
problem Lack of joint consideration of graph structure and node features in GNN interpretation.
method Model-agnostic framework using attention mechanism and sparsity regularization.
result GISST achieves superior node feature and edge explanation precision in synthetic and real-world datasets.
FEDS distills LIC model knowledge into a lightweight student for efficient compression.
problem Efficiently compress images with high performance and low resources.
method FEDS combines feature alignment and entropy-based loss for lightweight compression.
result Student model matches teacher's performance while reducing parameters and speeding up.
A new approach improves numerical tabular data imputation by addressing diffusion models' limitations.
problem Inaccurate and difficult training in numerical tabular data imputation.
method Kernelized Negative Entropy-regularized Wasserstein gradient flow Imputation (KnewImp) based on Wasserstein gradient flow (WGF) framework.
result KnewImp significantly outperforms existing methods in numerical tabular data imputation.
In medical real-world study (RWS), how to fully utilize the fragmentary and scarce information in model training to generate the solid diagnosis results is a challenging task. In this work, we introduce a novel multi-instance neural network, AMI-Net+, to train and predict from the incomplete and extremely imbalanced da…
The study improves representation learning bounds using data-dependent Gaussian mixtures.
problem Improving generalization in representation learning.
method Established bounds using relative entropy and MDL of latent variables.
result The approach significantly improves generalization over existing methods.