Task loss matching misrepresents similarity between neural network layers.
problem Measuring similarity between neural network layers using task loss matching.
method Task loss matching vs. direct matching; comparison with CCA and CKA.
result Direct matching provides a better similarity index than task loss matching.
Improves score estimation for noised targets using known clean scores.
problem Poor score estimation at low noise levels in Denoising Score Matching.
method Introduces Target Score Identity and Target Score Matching loss.
result Score estimates are more accurate at low noise levels.
A geometric theory explains loss functions for robust representation learning.
problem Treats robustness, domain adaptation, and sensor drift as separate literatures.
method Estimates covariance Sigma_task and uses it to pin Jacobian penalties.
result Proves optimality and necessity of range coverage for penalty matrices.
EGFs use ergodicity to simplify generative flows for easier training and imitation learning.
problem Challenges in training generative flows, especially in continuous settings and for imitation learning.
method EGFs leverage ergodicity to build simple flows with universality guarantees and tractable FM loss. They introduce a KL-weakFM loss for IL training without a separate reward model.
result EGFs simplify generative flow training and enable effective imitation learning.
New score matching method estimates local intrinsic dimension efficiently.
problem Quantifying the local intrinsic dimension of complex data.
method Denoising score matching loss and equivalent implicit score matching loss.
result Denoising score matching loss is a highly competitive and scalable LID estimator.
Paper tackles musical version matching at segment level using contrastive learning from weakly-labeled data.
problem Match musical versions at the segment level, not just tracks, with weak annotations.
method Proposes contrastive learning from weakly-labeled audio segments, using a new loss variant.
result Breakthrough performance in segment-level evaluation, outperforming state-of-the-art.
Despite being the standard loss function to train multi-class neural networks, the log-softmax has two potential limitations. First, it involves computations that scale linearly with the number of output classes, which can restrict the size of problems we are able to tackle with current hardware. Second, it remains unc…
Regularization and normalization have become indispensable components in training deep neural networks, resulting in faster training and improved generalization performance. We propose the projected error function regularization loss (PER) that encourages activations to follow the standard normal distribution. PER rand…
We present a powerful new loss function and training scheme for learning binary hash functions. In particular, we demonstrate our method by creating for the first time a neural network that outperforms state-of-the-art Haar wavelets and color layout descriptors at the task of automated scene matching. By accurately rel…
The paper justifies time-dependent loss reweighting schemes for flow matching and diffusion models.
problem Theoretical justification for time-dependent loss reweighting schemes in flow matching and diffusion models.
method Clarifies that the loss can depend on both time and state, and shows theoretical justification for time-dependent loss weighting schemes.
result Time-dependent loss weighting schemes are theoretically justified for Generator Matching and Edit Flows.
Paper analyzes proper losses and their performance in machine learning tasks.
problem Understanding the performance of estimators and forecasters in machine learning tasks.
method Analyzes surrogate regret and convergence rates for strictly proper losses.
result Strongly proper losses achieve the optimal convergence rate.
Proposes measures for uncertainty quantification using proper scoring rules.
problem Uncertainty quantification for prediction tasks.
method Decomposes proper scoring rules into divergence and entropy components, tailoring uncertainty quantification to specific tasks.
result Flexibility in uncertainty quantification improves performance in selective prediction and active learning.
A new approach to protein language models combines latent space prediction with masked language modeling.
problem Improving protein language models by predicting amino acid identities at masked positions.
method A variant of masked language modeling that predicts latent targets only at masked positions, retaining the MLM cross-entropy.
result The new approach outperforms pure masked language modeling on 11 out of 16 downstream tasks.
A new method for estimating uncertainties in neural ODEs without numerical integration.
problem Accurate estimation of predictive uncertainties in neural ODEs.
method Distributional Gradient Matching (DGM) algorithm that jointly trains a smoother and a dynamics model.
result Significantly more accurate predictions compared to traditional methods.
L2M learns to match distributions for domain adaptation without relying on hand-crafted priors.
problem Difficult to directly match cross-domain distributions; existing methods fail to uncover underlying relationships.
method Learning to Match (L2M) framework that learns distribution matching loss using a meta-network in a data-driven way.
result L2M outperforms state-of-the-art methods on public datasets and transfers from pneumonia to COVID-19 images.
Graph matching involves combinatorial optimization based on edge-to-edge affinity matrix, which can be generally formulated as Lawler's Quadratic Assignment Problem (QAP). This paper presents a QAP network directly learning with the affinity matrix (equivalently the association graph) whereby the matching problem is tr…
Closed-form flow matching yields similar performance to stochastic version, improving model performance.
problem Understanding why flow matching models generalize well.
method Empirical analysis and comparison of stochastic and closed-form flow matching losses.
result Closed-form flow matching can improve model performance.
GDT improves reinforcement learning by matching future state information efficiently.
problem Efficient learning of multi-task policies from trajectory data.
method Generalized Decision Transformer (GDT) for offline hindsight information matching.
result GDT enables effective offline multi-task state-marginal matching and imitation learning.
A new method for stochastic optimal control improves accuracy over existing techniques.
problem Improving the accuracy of stochastic optimal control for noisy systems.
method Stochastic Optimal Control Matching (SOCM) using Iterative Diffusion Optimization (IDO) with path-wise reparameterization trick.
result SOCM achieves lower error than existing techniques for three out of four control problems, sometimes by an order of magnitude.
PDDS samples from unnormalized densities using iterative particle scheme.
problem Sampling from unnormalized probability densities.
method Iterative particle scheme with novel score matching loss.
result Asymptotically consistent estimates for multimodal and high-dimensional tasks.
Unified calibration metrics improve forecast sharpness and accuracy.
problem Improving the sharpness of probabilistic forecasts while maintaining calibration.
method Kernel-based calibration metrics that unify and generalize existing methods for classification and regression.
result Enhanced calibration, sharpness, and decision-making across various tasks.
One of the fundamental tasks in understanding genomics is the problem of predicting Transcription Factor Binding Sites (TFBSs). With more than hundreds of Transcription Factors (TFs) as labels, genomic-sequence based TFBS prediction is a challenging multi-label classification task. There are two major biological mechan…
GMC benchmark isolates retrieval in Transformers, revealing max-margin alignment.
problem Understanding how Transformers develop match-and-copy behavior on natural data.
method Introducing Gaussian Match-and-Copy (GMC) as a minimalist benchmark.
result Gradient descent drives parameters to diverge while aligning with max-margin separator.
In this work we provide a theoretical framework for structured prediction that generalizes the existing theory of surrogate methods for binary and multiclass classification based on estimating conditional probabilities with smooth convex surrogates (e.g. logistic regression). The theory relies on a natural characteriza…
We consider the following conditional linear regression problem: the task is to identify both (i) a k-DNF condition c and (ii) a linear rule f such that the probability of c is (approximately) at least some given bound μ, and f minimizes the ℓp loss of predicting the target z in the distribution of …
Score matching offers efficient estimation for certain distributions.
problem Estimating probability distributions with intractable constants.
method Score matching as an alternative to maximum likelihood.
result Score matching is computationally and statistically efficient for certain distributions.
EpiMer merges models by solving Fréchet mean on a Riemannian manifold.
problem Integrating knowledge from multiple models without retraining.
method EpiMer casts model merging as solving the Fréchet mean on a Riemannian manifold, restricting computation to a low-rank subspace.
result EpiMer outperforms flat-geometry methods on image classification tasks.
Proposes a non-adversarial method for distribution matching.
problem Stability and optimization challenges in adversarial matching methods.
method Non-adversarial VAE-based matching method with alignment upper bounds.
result Demonstrates applicability of non-adversarial matching methods without modifying original architectures.
New method boosts performance of diffusion models on discrete data like natural language.
problem Performance of diffusion models on discrete data like natural language is poor.
method Proposes score entropy, a novel loss that extends score matching to discrete spaces.
result Significantly boosts performance on language modeling tasks.
When optimizing against the mean loss over a distribution of predictions in the context of a regression task, then even if there is a distribution of targets the optimal prediction distribution is always a delta function at a single value. Methods of constructing generative models need to overcome this tendency. We con…
Improved online convex optimization bounds between stochastic and adversarial settings.
problem Understanding optimization tasks that are neither i.i.d. nor fully adversarial.
method Establishing novel regret bounds exploiting smoothness of expected losses.
result Regret bounds improve on previous results by reducing dependence on maximum gradient length to variance of gradients.
New bounds for online convex optimization between stochastic and adversarial settings.
problem Understanding optimization tasks that are neither i.i.d. nor fully adversarial.
method Establishing novel regret bounds exploiting smoothness of expected losses.
result Regret bounds match expected rates in the fully i.i.d. case and gracefully deteriorate in the fully adversarial case.
LGD algorithm learns optimal hyperparameters for regression tasks.
problem Learning to learn hyperparameters for regression problems.
method Langevin Gradient Descent (LGD) approximates posterior distribution.
result Meta-learning optimal hyperparameters achieves Bayes' optimal solution.
The bipartite record linkage task consists of merging two disparate datafiles containing information on two overlapping sets of entities. This is non-trivial in the absence of unique identifiers and it is important for a wide variety of applications given that it needs to be solved whenever we have to combine informati…
The paper argues that uncertainty quantification in ML is application-specific and proposes a flexible family of measures.
problem The need for proper uncertainty quantification in machine learning for safety-critical applications.
method A flexible family of uncertainty measures tailored to specific applications, using proper scoring rules to control characteristics.
result Different uncertainty measures are more suitable for different tasks (e.g., selective prediction, out-of-distribution detection, active learning).
Maximum likelihood training improves the performance of score-based diffusion models.
problem Training score-based diffusion models with maximum likelihood.
method Trained by minimizing a weighted combination of score matching losses, with a specific weighting scheme that bounds negative log-likelihood.
result Maximum likelihood training improves the log-likelihood of score-based diffusion models across multiple datasets.
We propose an online convex optimization algorithm (RescaledExp) that achieves optimal regret in the unconstrained setting without prior knowledge of any bounds on the loss functions. We prove a lower bound showing an exponential separation between the regret of existing algorithms that require a known bound on the los…
This paper presents a learning method for convolutional autoencoders (CAEs) for extracting features from images. CAEs can be obtained by utilizing convolutional neural networks to learn an approximation to the identity function in an unsupervised manner. The loss function based on the pixel loss (PL) that is the mean s…
A new loss function ED simplifies training energy-based models without scores.
problem Training energy-based models is computationally expensive.
method Energy Discrepancy (ED) loss function that does not rely on scores or MCMC.
result ED effectively interpolates between score matching and negative log-likelihood.
Generative models learn distributions, new method finds inputs matching desired conditional distributions.
problem Designing inputs that produce specific conditional distributions, not just points.
method Conditional Distribution Matching (CDM) and MLGD-F algorithm.
result MLGD-F reliably recovers inputs matching diverse user-specified conditional distributions.
Hearing Aid (HA) algorithms need to be tuned ("fitted") to match the impairment of each specific patient. The lack of a fundamental HA fitting theory is a strong contributing factor to an unsatisfying sound experience for about 20% of hearing aid patients. This paper proposes a probabilistic modeling approach to the de…
OEUVRE estimates online loss with constant time and memory, outperforming other methods.
problem Accurately estimating expected loss in online learning.
method Recursive evaluation of each sample on current and previous models, using algorithmic stability for updates.
result Consistency, convergence rates, and concentration bounds proved for OEUVRE.
E-commerce sponsored search contributes an important part of revenue for the e-commerce company. In consideration of effectiveness and efficiency, a large-scale sponsored search system commonly adopts a multi-stage architecture. We name these stages as ad retrieval, ad pre-ranking and ad ranking. Ad retrieval and ad pr…
We introduce new families of Integral Probability Metrics (IPM) for training Generative Adversarial Networks (GAN). Our IPMs are based on matching statistics of distributions embedded in a finite dimensional feature space. Mean and covariance feature matching IPMs allow for stable training of GANs, which we will call M…
A new method for adapting to label shifts using class probability matching.
problem Adapting to label shifts where class probabilities differ between source and target domains.
method Class Probability Matching using Kernel Methods (CPMKM) framework.
result CPMKM outperforms existing methods on real datasets.
Heuristic weighting improves denoising score matching without requiring noise distribution assumptions.
problem Improving denoising score matching without assuming noise distribution.
method Demonstrated heteroskedasticity, derived optimal weighting functions, and provided theoretical and empirical comparisons.
result Heuristical weighting function can achieve lower variance than optimal weighting, facilitating more stable and efficient training.
This work classifies SOC loss functions based on their gradient properties.
problem Optimizing noisy systems in stochastic optimal control.
method Grouping loss functions into classes with the same gradient expectation.
result Different loss functions have the same optimization landscape but differ in gradient variance.
New method learns particle system potentials from unlabeled data.
problem Learning potentials of interacting particle systems from unlabeled data with trajectory information missing.
method Introduces a self-test loss function based on stochastic evolution equation.
result Method outperforms baseline methods in robust estimation of large, high-dimensional systems.