We study losses for binary classification and class probability estimation and extend the understanding of them from margin losses to general composite losses which are the composition of a proper loss with a link function. We characterise when margin losses can be proper composite losses, explicitly show how to determ…
This paper characterizes exp-concavity of proper composite losses and transforms mixable losses into exp-concave ones.
problem Understanding and transforming mixable losses into exp-concave ones for better online prediction strategies.
method Characterization of exp-concavity, mixability condition, and approximation approach for multi-class losses.
result Complete characterization of exp-concavity for proper composite losses and transformation of mixable losses into exp-concave ones.
We consider composite loss functions for multiclass prediction comprising a proper (i.e., Fisher-consistent) loss over probability distributions and an inverse link function. We establish conditions for their (strong) convexity and explore the implications. We also show how the separation of concerns afforded by using …
Extends loss function analysis to infinite dimensions for better machine learning.
problem Difficulty in separating machine learning problem study from loss function properties.
method Generalizes proper-composite representation to infinite dimensions, characterizing canonical link.
result Simple characterisation of canonical link in infinite dimensional setting.
The problem of bipartite ranking, where instances are labeled positive or negative and the goal is to learn a scoring function that minimizes the probability of mis-ranking a pair of positive and negative instances (or equivalently, that maximizes the area under the ROC curve), has been widely studied in recent years. …
Paper proposes fitting loss functions to data using source functions from information geometry.
problem Choosing appropriate loss functions for machine learning models.
method Introduces source functions from information geometry to fit loss functions to the domain at hand.
result Significant improvements over state-of-the-art methods in model training.
In this paper we construct proper biharmonic submanifolds into various types of ellipsoids. We also prove, in this context, some useful composition properties which can be used to produce large families of new proper biharmonic immersions.
Paper develops proper, lower-bounded losses for weakly supervised classification.
problem Weakly supervised classification with corrupted labels.
method Representation theorem for proper losses, derived condition for lower-boundedness, generalized logit squeezing.
result Proper and lower-bounded losses for weak-label learning.
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.
We study proper losses for discrete generative models without knowing the target distribution.
problem Evaluating generative models in the discrete setting without direct access to the target distribution.
method Define and construct black-box proper losses using statistical estimation theory.
result Black-box proper losses must be of polynomial form and involve more samples than the polynomial degree.
New findings on prime theta-curves with simple tangles.
problem Understanding prime theta-curves with specific unknotting numbers.
method Analyzing composite theta-curves and their components.
result Composite theta-curves with unknotting number one are prime.
This work broadens calibeating to various proper losses using Bregman divergence.
problem Calibration for a wide range of proper losses.
method Regret minimization and Bregman divergence approach.
result U-calibration results for a family of Tsallis losses with logarithmic regret and dimension independence.
This work generalizes calibeating for a broader range of proper losses using Bregman divergence.
problem Calibration for a wide range of proper losses beyond Brier and log loss.
method Regret minimization based on Bregman divergence for a family of proper losses.
result U-calibration results for a family of Tsallis losses with logarithmic regret and dimension independence.
Optimizing proper loss yields calibrated models under specific conditions.
problem Understanding when optimizing proper loss functions leads to calibrated predictions.
method Local optimality condition and Lipschitz functions.
result Predictors with local optimality are nearly calibrated and nearly locally optimal.
Study on proper learning under relaxed worst-case robust loss for VC classes.
problem Proper adversarially robust PAC learning under relaxed worst-case robust loss.
method Introduced a family of robust loss relaxations and showed their effectiveness for proper learnability.
result VC classes are properly PAC learnable with sample complexity close to standard PAC learning setup.
A new method learns proper multiclass losses and probabilities.
problem Learning proper multiclass losses for complex classification tasks.
method Extends monotonicity to multiclass problems using convex functions.
result Consistently outperforms natural multiclass baseline on up to 1,000 class datasets.
Edgeworth Accountant calculates privacy loss under differential privacy compositions efficiently.
problem Efficiently computing overall privacy loss under composition of private algorithms.
method Analytical approach using f f f -differential privacy framework and Edgeworth expansion. result Non-asymptotic ( ε , δ ) (ε, δ) ( ε , δ ) -differential privacy bounds with reduced computational cost. Optimal multiclass U-calibration error found to be Θ(√KT).
problem Online multiclass U-calibration with low regret for all bounded proper losses.
method Follow-the-Perturbed-Leader algorithm and lower bound construction.
result Optimal U-calibration error is Θ(√KT).
Paper simplifies DP composition for adaptive privacy budgets, enabling better privacy and accuracy in deep learning.
problem Tension between efficiency and flexibility in DP composition theorems.
method Rényi Differential Privacy (RDP) for adaptive privacy budgets, proving simpler composition theorem with smaller constants.
result Practical DP composition for adaptive privacy budgets, enabling better privacy and accuracy in deep learning.
Paper proposes a principled method to learn loss functions for supervised learning tasks.
problem Choosing an appropriate loss function for supervised learning tasks.
method The paper revisits and generalizes the SLIsotron algorithm using Bregman divergences.
result The BregmanTron algorithm learns both the loss and classifier, with convergence guarantees.
New approach for distributed online optimization of non-convex losses with sublinear regret.
problem Regret evaluation and consensus in distributed, multi-agent systems with non-convex losses.
method Composite regret metric and consensus-based online normalized gradient (CONGD) approach for pseudo-convex losses; offline optimization oracle for general non-convex losses.
result First sublinear regret bound for general distributed online non-convex learning.
In statistical analysis, measuring a score of predictive performance is an important task. In many scientific fields, appropriate scores were tailored to tackle the problems at hand. A proper score is a popular tool to obtain statistically consistent forecasts. Furthermore, a mathematical characterization of the proper…
Paper introduces arctan pinball loss for XGBoost quantile regression.
problem Efficiently predicting multiple quantiles with XGBoost.
method Smooth approximation of pinball loss for XGBoost, using arctan pinball loss.
result Arctan pinball loss reduces quantile crossings and improves efficiency.
New sparse GP model learns compositional kernels efficiently.
problem Learning accurate Gaussian Process models with complex kernel structures.
method MultiSVGP model with Horseshoe prior for kernel selection.
result Our model provides better fit and faster computation for large-scale data.
This paper advances FL algorithms for composite optimization and statistical recovery.
problem Federated learning optimization and statistical recovery in composite settings.
method Proposes Fast Federated Dual Averaging for strongly convex and smooth loss, and Multi-stage Federated Dual Averaging for restricted strongly convex and smooth loss.
result Establishes state-of-the-art iteration and communication complexity, and high probability complexity bound with linear speedup.
New algorithm reduces prediction errors across various loss functions.
problem Online forecasting algorithms' inability to adapt to different loss functions.
method Design of a novel Follow-the-Perturbed-Leader (FTPL) algorithm with self-concordant noise.
result Simultaneously achieves i l d e O ( T ) ilde O(\sqrt{T}) i l d e O ( T ) regret for bounded proper losses and O ( log T ) O(\log T) O ( log T ) regret for bounded smooth proper losses. The paper decomposes probabilistic scores into reliability, uncertainty, and information loss.
problem Understanding the reliability and uncertainty of probabilistic predictions.
method Developed decomposition identities for proper losses, quantifying reliability, residual uncertainty, and information gain.
result A three-term identity for classification scores, revealing miscalibration, grouping term, and feature-level uncertainty.
New filters match advanced composition for adaptive privacy, with practical constants.
problem Limitations of existing adaptive composition methods.
method Constructed new filters and odometers that match advanced composition rates, including constants.
result Achieved fully adaptive privacy with practical filters and odometers.
Paper extends FFT-based differential privacy method to heterogeneous compositions.
problem Computing accurate differential privacy guarantees for mixed mechanisms.
method Uses Fast Fourier Transform (FFT) for error analysis and parameter selection.
result Provides tighter bounds for heterogeneous compositions compared to homogeneous cases.
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.
Analysis of a stochastic system showing convergence to an averaged model with Gaussian deviations.
problem Convergence analysis of a perturbed compositional gradient flow system.
method Separation of scales and averaging principle applied to stochastic differential equations.
result The slow motion of the system can be approximated by a standard perturbed gradient flow or SCGD algorithm.
Unified framework for estimating density ratios across multiple distributions.
problem Binary density ratio estimation for multiple distributions.
method Unified framework based on Bregman divergence minimization.
result Generalization of binary DRE methods to multiple distributions.
AutoBayes simplifies variational inference by composing models and optimizing them.
problem Generalized variational inference complexities and optimization challenges.
method Compositional framework exploiting chain rules for automatic differentiation.
result Optimized models and parameterized statistical games can be locally optimized.
A new privacy accountant for Gaussian differential privacy measures individual privacy losses.
problem Bounding differential privacy loss for each participant in data analysis.
method Developed a privacy accountant for adaptive compositions of randomised mechanisms using Gaussian differential privacy.
result Provided optimal bounds for the Gaussian mechanism and constructed an approximative individual privacy accountant.
New algorithms minimize dynamic regret for strongly convex losses.
problem Minimizing dynamic regret for strongly convex losses.
method Developed Strongly Adaptive algorithms exploiting KKT conditions.
result Achieved near optimal dynamic regret of O ( d 1 / 3 n 1 / 3 e x t T V [ u 1 : n ] 2 / 3 ∨ d ) O(d^{1/3} n^{1/3} ext{TV}[u_{1:n}]^{2/3} \vee d) O ( d 1/3 n 1/3 e x t T V [ u 1 : n ] 2/3 ∨ d ) . A new model for data with zeros or missing values.
problem Data with excess zeros or missing values.
method Composite loss framework for low-rank modeling, combining generalized low-rank and hurdle methods.
result Demonstrated on a manufacturing data set and applied to missing value imputation.
Since its inception, the modus operandi of multi-task learning (MTL) has been to minimize the task-wise mean of the empirical risks. We introduce a generalized loss-compositional paradigm for MTL that includes a spectrum of formulations as a subfamily. One endpoint of this spectrum is minimax MTL: a new MTL formulation…
Study non-oblivious adversarial bandits with delayed feedback and propose algorithms with improved regret bounds.
problem Adversarial bandit problem with delayed, composite anonymous feedback.
method Propose wrapper algorithm for non-oblivious delay setting, achieving o ( T ) o(T) o ( T ) policy regret. result Achieve o ( T ) o(T) o ( T ) policy regret for many adversarial bandit problems with bounded memory loss sequences. Extends knockoff filter for composite null hypotheses in variable selection.
problem Handling composite null hypotheses in variable selection.
method Developed two methods for composite inference with knockoffs: S-OLS and FRPP.
result Proposed heuristic variants of S-OLS outperforming BH procedure for composite nulls.
Generalization error defines the discriminability and the representation power of a deep model. In this work, we claim that feature space design using deep compositional function plays a significant role in generalization along with explicit and implicit regularizations. Our claims are being established with several im…
Improves generative models for cost-sensitive decisions.
problem Generative models lack awareness of decision costs.
method Integrates a decision loss into the training objective.
result Improves cost-sensitive forecast accuracy.
New method improves decision-making accuracy without complex calculations.
problem Improving decision-making accuracy in machine learning.
method Introducing a new measure called calibration decision loss ( C D L K \mathsf{CDL}_K CDL K ) for structured families of post-processing functions. result Proves upper and lower bounds for natural classes K K K of post-processing functions. 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.
Unified analysis of multi-task functional linear regression with manifold and composite penalties.
problem Estimating slope functions from functional data with multi-task learning.
method Penalized splines with manifold constraint and composite quadratic penalty.
result Unified convergence upper bound and phase transition behaviors for estimators.
Improved subgradient method tackles ill-conditioned composite optimization problems.
problem Slow convergence of subgradient method for composite optimization problems.
method Preconditioned subgradient method with Levenberg-Marquardt approach.
result Linear convergence rate for composite optimization problems under mild conditions.
Inverse depth scaling found in LLMs due to similar layers averaging error.
problem Understanding how depth affects loss in large language models.
method Analysis of LLMs and toy residual networks.
result Loss scales inversely proportional to depth in LLMs.
The intuition of risk is based on two main concepts: loss and variability. In this paper, we present a composition of risk and deviation measures, which contemplate these two concepts. Based on the proposed Limitedness axiom, we prove that this resulting composition, based on properties of the two components, is a cohe…
New findings on adversarial training robustness.
problem How to create a resource-bounded adversary that severely damages learning.
method Formalizes the problem with proper losses and a central measure of 'harmfulness'. Identifies a sufficient property for adversaries to be detrimental.
result Optimizing a central measure for a subset of proper losses is independent of the loss and involves optimal transport.