Paper develops new methods for binary classification with complex performance measures.
problem Complex performance measures in binary classification are not decomposable and require new theoretical and methodological developments.
method Identifies Karmic and threshold-quasi-concavity properties, and develops a computationally practical plug-in classifier.
result Bayes optimal classifier is a threshold function of conditional probability, leading to practical classification error analysis.
Proposes Neural Complexity (NC) for predicting and explaining generalization in deep neural networks.
problem Challenges in specifying a suitable complexity measure for deep neural networks to predict and explain generalization.
method A meta-learning framework that learns a scalar complexity measure through interactions with many heterogeneous tasks.
result Trained NC model can be added to standard training loss to regularize any task learner.
Study shows simple vector quantization measures correlate with deep learning generalization.
problem Understanding and predicting generalization in deep learning models.
method Applying complexity measures from approximation and information theory to deep learning features.
result Simple vector quantization measures correlate well with generalization performance in deep learning.
Deep learning models can initially degrade in performance as they grow larger, then improve.
problem Performance degradation of larger models and more data.
method Defined effective model complexity and identified double descent phenomenon.
result Increasing model size and data can initially hurt performance, contrary to intuition.
Deep kernel learning for complex function modeling.
problem Modeling complex functions with line integral measurements.
method Gaussian process with neural networks for line integral data.
result Improved performance in computed tomography reconstruction.
New topological complexity measures for neural networks.
problem Measuring complexity of neural network functions.
method Generalized piecewise-linear Morse theory applied to ReLU networks.
result Local complexity can be arbitrarily high.
A new method, Multi-Label Deep Forest, tackles multi-label learning problems.
problem Leveraging label correlations in multi-label learning models.
method Designs a deep forest framework with two mechanisms: measure-aware feature reuse and measure-aware layer growth.
result Outperforms compared methods on six measures across benchmark datasets.
Unified framework for learning quantum models from limited measurements.
problem Sample complexity and measurement shots in classical learning of quantum models.
method Unified learning framework considering probabilistic quantum measurements.
result Asymmetrical effects and interplay of sample size and measurement shots on learning performance.
Consistent algorithms for multiclass learning with complex metrics and constraints.
problem Learning with complex performance metrics and constraints.
method General framework for designing consistent algorithms by viewing the problem as an optimization over feasible confusion matrices.
result Rates of convergence to the optimal (feasible) classifier, showing asymptotic consistency.
Protein function prediction is the important problem in modern biology. In this paper, the un-normalized, symmetric normalized, and random walk graph Laplacian based semi-supervised learning methods will be applied to the integrated network combined from multiple networks to predict the functions of all yeast proteins …
Paper develops a fractal dimension-based generalization measure.
problem Developing a robust generalization measure for machine learning models.
method Analyzes decision boundaries using fractal dimension concept.
result Developed a generalization measure based on fractal dimension.
NeurIPS 2020 competition seeks to predict deep learning generalization.
problem Understanding and predicting generalization in deep learning models.
method Propose complexity measures to accurately predict generalization performance.
result A robust complexity measure could improve deep learning reliability.
A new perceptual adjustment query for metric learning reduces complexity in high-dimensional data.
problem Metric learning in high-dimensional data with limited human feedback.
method Inverted measurement scheme and two-stage estimator for PAQs.
result Sample complexity guarantees for the two-stage estimator of metric learning from PAQs.
Study improves material similarity measures considering distinctiveness.
problem Improving similarity measures for materials science applications.
method Used machine learning techniques with specific descriptors and kernels.
result Minimizing loss of distinctiveness improves prediction accuracy.
New tools quantify deep generative models' performance.
problem Measuring the quality-diversity trade-off in deep generative models.
method Established non-asymptotic bounds on sample complexity and introduced frontier integrals.
result Smoothed estimators improve convergence rates of divergence frontiers.
New method measures generalizability of deep neural networks based on decision boundary complexity.
problem Lack of generalization methods for deep neural networks.
method Created Decision Boundary Complexity (DBC) score to measure DNN complexity.
result Simpler decision boundaries lead to better generalizability, supporting Occam's Razor.
The paper analyzes tensor recovery from symmetric rank-one measurements using information theory.
problem Recovering tensors with low symmetric rank from symmetric rank-one measurements.
method Covering numbers argument, Carbery-Wright inequality, orthogonal polynomials, Fano's inequality.
result Near-optimal sample complexity bounds for log-concave distributions.
New algorithms improve contextual bandit performance by adapting to problem difficulty.
problem Improving contextual bandit performance on problems with varying difficulty.
method Introducing complexity measures and oracle-efficient algorithms.
result Achieves optimal instance-dependent regret bounds for rich policy classes.
In data science, it is often required to estimate dependencies between different data sources. These dependencies are typically calculated using Pearson's correlation, distance correlation, and/or mutual information. However, none of these measures satisfy all the Granger's axioms for an "ideal measure". One such ideal…
This paper tackles denoising of complex measures using optimal transport and curvature analysis.
problem Denoising of complex, possibly non-log-concave measures.
method Score function and optimal transport theory to revert Langevin diffusion chains.
result The difficulty of denoising depends on the curvature complexity of the initial measure at specific SNR scales.
AI benchmarks evaluate football team performance using generative models.
problem Evaluating human performance in complex interactive tasks is error-prone and unreliable.
method Trained Conditional VRNN Model on player and ball tracking data to imitate and predict team interactions.
result Trained model as a useful benchmark for evaluating team performance in football.
Hadamard Wirtinger Flow recovers sparse signals from fewer measurements.
problem Reconstructing sparse signals from magnitude-only measurements.
method Gradient descent with Hadamard parametrization (HWF).
result A single step of HWF recovers support from k(xmax∗)−2 samples. New insights into neural network complexity reveal better generalization performance.
problem Mysterious generalization in deep models despite high parameter counts.
method Effective dimensionality as a measure of parameter space complexity.
result Double descent behavior in generalization as a function of parameters explained.
Improved sampling strategy reduces Fourier measurements for neural network signals.
problem Efficiently sampling signals from neural networks with random Fourier matrices.
method Model-adapted sampling strategy with improved sample complexity.
result Reduced sample complexity from O(kdnα∞²) to O(kdα²₂) measurements.
We present a framework and analysis of consistent binary classification for complex and non-decomposable performance metrics such as the F-measure and the Jaccard measure. The proposed framework is general, as it applies to both batch and online learning, and to both linear and non-linear models. Our work follows recen…
Some neural network modules are more critical to performance than others.
problem Understanding why some neural network architectures generalize better than others.
method Introduced module criticality, a measure based on the shape of loss valleys.
result Module criticality explains superior generalization performance of some architectures.
New risk bound derived for multi-category margin classifiers.
problem Guaranteed risk dependency on categories, sample size, and margin parameter.
method Derived a new risk bound using Rademacher complexity and chaining method.
result Improved dependency on categories over state of the art.
This work introduces a method to learn dynamical systems from noisy sensor measurements using multiple shooting.
problem Learning dynamical systems from noisy sensor measurements is challenging due to system instability.
method A scalable method based on multiple shooting.
result Robust learning of latent representations of dynamical systems from noisy measurements.
Defines a new metric to measure importance of predictors in complex machine learning models.
problem Measuring importance of predictors in black box machine learning models.
method Introduces a new metric, GVIM, based on true conditional expectation functions and causal interpretation.
result The GVIM can be represented as a function of Conditional Average Treatment Effect (CATE), providing a causal interpretation.
Paper proposes method for optimal control of unknown systems with latent states.
problem Jointly estimating dynamics and latent states in systems with unmeasurable states.
method Combination of particle Markov chain Monte Carlo methods and scenario theory.
result Probabilistic performance guarantees for optimal input trajectories.
A new complexity measure MDL-COMP for overparameterized models improves generalization performance.
problem Complexity measures based on Rissanen's MDL principle are not well-suited for overparameterized models.
method Developed a novel MDL-based complexity (MDL-COMP) for overparameterized models, defined via an optimality criterion over Ridge estimators.
result MDL-COMP scales linearly with d when d<n, but exponentially smaller for d>n; it upper bounds in-sample MSE. The article introduces gamma-Psi-dimensions for margin multi-category classifiers.
problem Margin multi-category classifiers' generalization performance under minimal learnability hypotheses.
method Derives gamma-Psi-dimensions, handles capacity measures, and establishes upper bounds on metric entropies and Rademacher complexity.
result Gamma-Psi-dimensions improve over fat-shattering dimension and offer a promising alternative for multi-class to binary transitions.
New complexity measures explain overparameterized models' surprising performance.
problem Understanding why overparameterized models generalize well despite fitting training data.
method Reinterpreting classical degrees of freedom in a random-X setting.
result Random-X prediction error better explains generalization in complex models.
Compression-based similarity measures are effectively employed in applications on diverse data types with a basically parameter-free approach. Nevertheless, there are problems in applying these techniques to medium-to-large datasets which have been seldom addressed. This paper proposes a similarity measure based on com…
Diffusion maps help learn complex quantum phase transitions from data.
problem Learning quantum phase transitions from experimental data is challenging.
method Diffusion maps for nonlinear dimensionality reduction and spectral clustering.
result Diffusion maps can learn complex phase transitions unsupervised.
Method recovers complex-valued signals from speckle-noised measurements.
problem Recovering complex-valued signals from speckle-noised measurements.
method Bagged Deep Image Priors integrated with projected gradient descent and Newton-Schulz algorithm.
result Achieves state-of-the-art performance in MSE reduction.
We consider a non-stationary variant of a sequential stochastic optimization problem, in which the underlying cost functions may change along the horizon. We propose a measure, termed variation budget, that controls the extent of said change, and study how restrictions on this budget impact achievable performance. We i…
New methods evaluate data representations by complexity of low-loss predictor learning.
problem Evaluating quality of data representations for downstream tasks.
method Surplus Description Length (SDL) and ε Sample Complexity (εSC) methods.
result Methods measure the information needed to approximate optimal predictor up to specified tolerance.
We information-theoretically reformulate two measures of capacity from statistical learning theory: empirical VC-entropy and empirical Rademacher complexity. We show these capacity measures count the number of hypotheses about a dataset that a learning algorithm falsifies when it finds the classifier in its repertoire …
In this short note, we provide a sample complexity lower bound for learning linear predictors with respect to the squared loss. Our focus is on an agnostic setting, where no assumptions are made on the data distribution. This contrasts with standard results in the literature, which either make distributional assumption…
Networked sensing, where the goal is to perform complex inference using a large number of inexpensive and decentralized sensors, has become an increasingly attractive research topic due to its applications in wireless sensor networks and internet-of-things. To reduce the communication, sensing and storage complexity, t…
Paper introduces a new, tractable measure of model complexity.
problem Need for a reliable measure of model complexity.
method Mathematically rigorous measure based on gradient similarities.
result Generalizes to various model types and insights into double descent.
A method to assess variable importance in complex predictive models.
problem Assessing the importance of variables in complex predictive models.
method Assigning relevance measures to each variable by comparing predictions with a ghost variable and analyzing joint effects.
result The method provides insights into variable importance and joint effects not available with other methods.
Measuring conditional dependencies among the variables of a network is of great interest to many disciplines. This paper studies some shortcomings of the existing dependency measures in detecting direct causal influences or their lack of ability for group selection to capture strong dependencies and accordingly introdu…
Paper combines DQN and return-based RL for improved policy performance.
problem Improving policy performance in reinforcement learning.
method Integrates DQN and return-based reinforcement learning, introduces two measurements to quantify policy discrepancy.
result The proposed measurements accurately express trace coefficient and improve approximation to return.
Develops a complexity measure for neural networks based on quantum statistical mechanics.
problem Understanding the relationship between neural network structure and generalization ability.
method Introduces Periodic Spectral Ergodicity (PSE) and cascading PSE (cPSE) to quantify neural network complexity.
result Demonstrates the effectiveness of cPSE in quantifying complexity and guiding NAS.
In this paper, we generalize Huber's criterion to multichannel sparse recovery problem of complex-valued measurements where the objective is to find good recovery of jointly sparse unknown signal vectors from the given multiple measurement vectors which are different linear combinations of the same known elementary vec…
New measure quantifies task difficulty for machine learning models.
problem Quantifying the inherent difficulty of machine learning tasks.
method Inductive bias complexity measure.
result Tasks requiring generalization over many dimensions are more difficult.