We study consistency of learning algorithms for a multi-class performance metric that is a non-decomposable function of the confusion matrix of a classifier and cannot be expressed as a sum of losses on individual data points; examples of such performance metrics include the macro F-measure popular in information retri…
On-device research index
arXiv research
A locally-built, LLM-digested index of recent arXiv papers in quant finance, geometry/topology, and statistical ML — keyword search served straight from SQLite on this machine.
169,341 papers · 148 categories
Trend · papers per month
48 results for “non-decomposable metrics”
We provide a general theoretical analysis of expected out-of-sample utility, also referred to as decision-theoretic classification, for non-decomposable binary classification metrics such as F-measure and Jaccard coefficient. Our key result is that the expected out-of-sample utility for many performance metrics is prov…
A new method for optimizing non-decomposable metrics with constraints.
problem Optimizing complex machine learning objectives with thresholded constraints.
method Formulate rate-constrained optimization using the Implicit Function theorem and solve with gradient-based methods.
result Demonstrated effectiveness over existing methods on benchmark datasets.
New method for recommending labels with missing data, improving performance metrics.
problem Recommending relevant labels for data points with missing labels and non-decomposable metrics.
method Proposes a framework to devise a regularized objective function and threshold for non-decomposable metrics like F1 measure.
result Bounded regret or generalization error in F1 measure and other metrics, even with missing labels.
AP-Perf integrates custom metrics into neural networks.
problem Incorporating non-decomposable performance metrics into differentiable learning.
method Adversarial prediction framework optimizing metrics in worst-case distribution.
result Demonstrated effectiveness on various classification tasks.
SelMix fine-tunes pre-trained models to optimize non-decomposable objectives.
problem Optimizing non-decomposable performance measures for practical applications.
method Selective mixup fine-tuning of pre-trained models.
result SelMix significantly improves performance for various non-decomposable objectives.
Paper tackles noisy labels for non-decomposable performance measures.
problem Learning from noisy labels for non-decomposable performance measures.
method Designs algorithms for multiclass non-decomposable performance measures using Frank-Wolfe and Bisection methods, corrected for class-conditional noise.
result Noise-corrected algorithms are Bayes consistent, converging to optimal performance.
New algorithm improves convergence of AUC maximization.
problem Optimizing AUC for imbalanced classes with stochastic methods.
method Variance Reduced Stochastic Proximal Algorithm for AUC Maximization (VRSPAM).
result VRSPAM converges faster than previous methods.
New deep learning algorithms optimize non-decomposable measures.
problem Optimizing complex, non-decomposable performance measures.
method Directly training deep neural networks with task-specific loss functions.
result Significantly faster and more stable convergence compared to standard methods.
A framework for multiclass/multioutput classification metrics, revealing geometric insights and consistency.
problem Developing robust metrics for multiclass/multioutput classification problems.
method Proposes a framework for constructing and analyzing multiclass/multioutput classification metrics, revealing geometric insights and characterizing averaging methodologies.
result Plug-in estimator based on the characterization is consistent and easily implemented.
Develops a new minimax probability machine for imbalanced classification tasks.
problem Imbalanced classification tasks with non-decomposable performance measures.
method Derives an equivalent form of the MPMF model for solving linear and nonlinear classifiers.
result Demonstrates the effectiveness of the new model on real-world datasets.
A new method, VIF, calculates influence for non-decomposable losses efficiently.
problem Efficiently calculating influence for complex machine learning models with non-decomposable losses.
method Revisiting influence function from robust statistics, proposing Versatile Influence Function (VIF) for any non-decomposable loss.
result VIF method is up to 10^3 times faster than brute-force methods and closely matches influence results.
Modern applications in sensitive domains such as biometrics and medicine frequently require the use of non-decomposable loss functions such as precision@k, F-measure etc. Compared to point loss functions such as hinge-loss, these offer much more fine grained control over prediction, but at the same time present novel c…
Unified approach for fair classification with overlapping groups.
problem Ensuring fairness across multiple overlapping groups in prediction problems.
method Probabilistic population analysis leading to Bayes-optimal classifier, unifying existing methods.
result Outperforms baselines in fairness-performance tradeoff on real datasets.
DeepTopPush improves accuracy at the top for complex classification tasks.
problem Minimizing irrelevant samples above a threshold in binary classification.
method Proposes a new method for end-to-end training of deep networks to minimize loss at the top.
result Demonstrates excellent performance on visual recognition and real-world applications.
We optimize rank-based metrics using blackbox differentiation.
problem Challenges in directly optimizing rank-based metrics due to their non-differentiable and non-decomposable nature.
method Efficient, theoretically sound, and general method for differentiating rank-based metrics with mini-batch gradient descent.
result Competitive performance on standard image retrieval datasets and improved performance on object detectors.
Houdini generates adversarial examples for deep structured prediction models.
problem Evaluating and improving the robustness of deep learning models, especially for non-decomposable tasks.
method Introduces Houdini, a flexible approach for generating adversarial examples tailored to the final performance measure of the task.
result Houdini achieves higher success rates in generating adversarial examples compared to traditional methods, using less perceptible perturbations.
New algorithm optimizes complex metrics in online learning.
problem Optimizing non-decomposable metrics in sequential learning.
method General online algorithm for various metrics.
result Achieves regret for concave and smooth metrics.
Paper proposes a method to minimize non-differentiable loss functions.
problem Minimizing non-differentiable and non-decomposable loss functions.
method Learn smooth relaxations of true losses through surrogate neural networks, then optimize jointly with the prediction model.
result Empirical results show the efficiency of learning surrogate losses.
A new game-theoretic approach optimizes complex rate metrics.
problem Optimizing non-decomposable performance metrics and rate constraints.
method Extending two-player game approaches to a three-player game, seeking equilibrium.
result Generalizes and improves upon existing algorithms for constrained optimization.
New framework for optimizing machine learning risks.
problem Optimizing non-decomposable machine learning objectives.
method Empirical X-risk minimization (EXM) framework with algorithmic techniques.
result Developed algorithms for solving EXM with smooth non-convex objectives.
Bayesian method learns Gaussian graphical models without decomposability constraints.
problem Learning non-decomposable Gaussian graphical models efficiently and accurately.
method Fractional pseudo-likelihood and sparsity-inducing prior.
result Consistent estimator of graph structure for high-dimensional data.
Vision problems ranging from image clustering to motion segmentation to semi-supervised learning can naturally be framed as subspace segmentation problems, in which one aims to recover multiple low-dimensional subspaces from noisy and corrupted input data. Low-Rank Representation (LRR), a convex formulation of the subs…
Unified framework for scalable optimization of ranking-based objectives.
problem Scalability issues in optimizing ranking-based performance metrics.
method Unified framework using building block bounds for scalable optimization.
result Substantial improvement in performance over accuracy-objective baseline.
Framework for consistent binary classification with complex metrics.
problem Consistent binary classification for non-decomposable metrics like F-measure and Jaccard.
method General framework for batch and online learning, applies to linear and non-linear models. Uses thresholding and normalized gradient ascent for threshold estimation.
result Simple normalized gradient ascent updates for threshold estimation, with finite-sample regret analysis.
Modern classification problems frequently present mild to severe label imbalance as well as specific requirements on classification characteristics, and require optimizing performance measures that are non-decomposable over the dataset, such as F-measure. Such measures have spurred much interest and pose specific chall…
Introduces robust and decomposable AP for image retrieval.
problem Challenges in training deep neural networks with AP.
method Differentiable rank approximation and loss function design.
result ROADMAP outperforms AP approximation methods and deep models.
New methods link Legendrian satellites to Lagrangian cobordisms.
problem Understanding relations between Legendrian and Lagrangian knots.
method Constructing Lagrangian concordances through satellite operations.
result Maximum Thurston-Bennequin number restricts Legendrian satellite Lagrangian sliceness.
Using gauge theory for Spin(7)-manifolds of dimension 8, we develop a procedure, called Spin-rotation, which transforms a (stable) holomorphic structure on a vector bundle over a complex torus of dimension 4 into a new holomorphic structure over a different complex torus. We show non-trivial examples of this procedure …
We introduce Clique Matrices as an alternative representation of undirected graphs, being a generalisation of the incidence matrix representation. Here we use clique matrices to decompose a graph into a set of possibly overlapping clusters, de ned as well-connected subsets of vertices. The decomposition is based on a s…
Range penalization enhances statistical accuracy and resource efficiency in federated learning.
problem Statistical accuracy and resource efficiency in federated learning.
method Range regularization and polar clustering.
result Enhanced statistical accuracy and reduced iteration complexity.
Backdrop uses dropout-like masking in backpropagation for multi-scale data.
problem Improving generalization in multi-scale data.
method Inserting masking layers after convolutional layers to mask backward gradients.
result Backdrop leads to significant improvements in generalization.
We prove that for any open Riemann surface natural number non-constant harmonic map and holomorphic 2-form on there exists a weakly complete harmonic map with Hopf differential and In particular,…
Graphical models provide powerful tools to uncover complicated patterns in multivariate data and are commonly used in Bayesian statistics and machine learning. In this paper, we introduce the R package BDgraph which performs Bayesian structure learning for general undirected graphical models (decomposable and non-decom…
This paper tackles unbiased loss functions for multilabel classification with missing labels.
problem Missing labels in multilabel classification tasks, especially in extreme multi-label classification (XMC).
method Derives unbiased estimators for multilabel reductions, including non-decomposable ones, and addresses increased variance with convex upper-bounds.
result Switching to unbiased estimators can alter the bias-variance trade-off and may require stronger regularization.
Develops gradient boosting for multi-label classification.
problem Lack of customizable learning algorithms for multi-label classification.
method Generalizes gradient boosting to multi-output problems and proposes an algorithm for learning multi-label classification rules.
result Ability to minimize both decomposable and non-decomposable loss functions.
The study connects fairness constraints with optimal transport to derive new insights in classification.
problem Ensuring fairness in classification models without sacrificing performance.
method Using Wasserstein barycenters and optimal transport, the study characterizes optimal classification functions under fairness constraints.
result Maximizing fairness under demographic parity is equivalent to solving a regression problem.
FNNC framework ensures fairness in neural networks through convex surrogates.
problem Ensuring fairness in neural network classification models.
method FNNC framework uses neural networks to include fairness constraints in the loss function and optimizes using mini-batch stochastic gradient descent.
result FNNC achieves fairness while maintaining high accuracy, as shown by experiments.
This paper explores methods for combining predictions in multilabel classification.
problem Lack of formal framework for aggregation in multilabel ensembles.
method Introduces two approaches: 'predict then combine' (PTC) and 'combine then predict' (CTP).
result Standard voting techniques are outperformed by tailored instantiations of CTP and PTC.
New method reduces bias and variance in OPE for large action spaces.
problem High bias and variance in OPE for large, combinatorial action spaces.
method Factored action spaces and decomposed importance sampling.
result Decomposed IS estimators have less variance than non-decomposed versions.
Paper tackles non-convex inf-projection problems with stochastic optimization.
problem Non-convex and possibly non-smooth inf-projection minimization problems.
method Developed stochastic algorithms for finding (nearly) stationary solutions.
result Established first-order convergence for non-convex inf-projection problems.
Bayesian learning for forests and trees improves graph detection and structure learning.
problem Learning graph structures in non-decomposable graphs.
method Adapted MCMC and SSS algorithms for forests and trees, using the Chow-Liu algorithm and Matrix Tree Theorem.
result SSS with trees or forests outperforms SSS with decomposable graphs in certain cases.
Improved performance of factorized neural layers through spectral initialization and Frobenius decay.
problem Improving the performance of factorized neural layers in various deep learning contexts.
method Spectral initialization and Frobenius decay for initialization and regularization.
result Spectral initialization and Frobenius decay lead to improved performance across multiple deep learning settings.
Federated Deep AUC Maximization for Heterogeneous Data with a Constant Communication Complexitycs.LG
Improved FDAM algorithms for heterogeneous data with constant communication complexity.
problem Maximizing AUC for imbalanced data classification in federated learning.
method Solving non-convex strongly-concave min-max formulation in a distributed fashion.
result Communication complexity is a constant, independent of number of machines and accuracy level.
Understanding Value Decomposition Algorithms in Deep Cooperative Multi-Agent Reinforcement Learningcs.LG
The paper explores when and why value decomposition algorithms work in cooperative multi-agent reinforcement learning.
problem The applicability and convergence properties of value decomposition algorithms in cooperative multi-agent reinforcement learning are unclear.
method The paper introduces decomposable games and proves that applying the multi-agent fitted Q-Iteration algorithm leads to an optimal Q-function in these games.
result The paper offers theoretical insights into when and why value decomposition algorithms converge in cooperative multi-agent reinforcement learning.
FeDXL tackles federated learning for X-risk optimization.
problem Optimizing a family of X-risks with federated learning, where existing algorithms are not applicable.
method Active-passive decomposition framework, federated averaging and merging, novel theoretical analysis.
result FeDXL algorithms for linear and nonlinear are developed, with established complexities and improved performance.
Survey on the Canonical Metrics on the Teichmüller Spaces and the Moduli Spaces of Riemann Surfacesmath.DG
This thesis surveys various metrics on Riemann surface spaces.
problem Various metrics on Riemann surface spaces.
method Survey of metrics and their properties.
result Equivalence of Kähler-Einstein metric to Teichmüller metric.
Study on conditions for Randers metrics to be of constant Ricci curvature.
problem Conditions for Randers metrics to be of constant Ricci curvature.
method Analysis of sufficient and necessary conditions for Randers metrics with and without strong convexity.
result Classification of Randers metrics with and .