Loss Functions and Operators Generated by f-Divergences

• URL: http://arxiv.org/abs/2501.18537v1
• Date: Thu, 30 Jan 2025 18:06:18 GMT
• Title: Loss Functions and Operators Generated by f-Divergences
• Authors: Vincent Roulet, Tianlin Liu, Nino Vieillard, Michael E. Sander, Mathieu Blondel,
• Abstract summary: We propose to construct new convex loss functions based on $f$-divergences. By analogy with the logistic loss, the loss function generated by an $f$-divergence is associated with an operator, that we dub $f$-softargmax. One of the goals of this paper is to determine the effectiveness of loss functions beyond the classical cross-entropy in a language model setting.
• Score: 21.58093510003414
• License:
• Abstract: The logistic loss (a.k.a. cross-entropy loss) is one of the most popular loss functions used for multiclass classification. It is also the loss function of choice for next-token prediction in language modeling. It is associated with the Kullback--Leibler (KL) divergence and the softargmax operator. In this work, we propose to construct new convex loss functions based on $f$-divergences. Our loss functions generalize the logistic loss in two directions: i) by replacing the KL divergence with $f$-divergences and ii) by allowing non-uniform reference measures. We instantiate our framework for numerous $f$-divergences, recovering existing losses and creating new ones. By analogy with the logistic loss, the loss function generated by an $f$-divergence is associated with an operator, that we dub $f$-softargmax. We derive a novel parallelizable bisection algorithm for computing the $f$-softargmax associated with any $f$-divergence. On the empirical side, one of the goals of this paper is to determine the effectiveness of loss functions beyond the classical cross-entropy in a language model setting, including on pre-training, post-training (SFT) and distillation. We show that the loss function generated by the $\alpha$-divergence (which is equivalent to Tsallis $\alpha$-negentropy in the case of unit reference measures) with $\alpha=1.5$ performs well across several tasks.

Related papers

• Any-stepsize Gradient Descent for Separable Data under Fenchel--Young Losses [17.835960292396255]
  We show arbitrary-stepsize gradient convergence for a general loss function based on the framework of emphFenchel--Young losses. We argue that these better rate is possible because of emphseparation margin of loss functions, instead of the self-bounding property.
  arXiv  Detail & Related papers  (2025-02-07T12:52:12Z)
• $α$-Divergence Loss Function for Neural Density Ratio Estimation [0.0]
  Density ratio estimation (DRE) is a fundamental machine learning technique for capturing relationships between two probability distributions. Existing methods face optimization challenges, such as overfitting due to lower-unbounded loss functions, biased mini-batch gradients, vanishing training loss gradients, and high sample requirements for Kullback-Leibler (KL) divergence loss functions. We propose a novel loss function for DRE, the $alpha$-divergence loss function ($alpha$-Div), which is concise but offers stable and effective optimization for DRE.
  arXiv  Detail & Related papers  (2024-02-03T05:33:01Z)
• Universal Online Learning with Gradient Variations: A Multi-layer Online Ensemble Approach [57.92727189589498]
  We propose an online convex optimization approach with two different levels of adaptivity. We obtain $mathcalO(log V_T)$, $mathcalO(d log V_T)$ and $hatmathcalO(sqrtV_T)$ regret bounds for strongly convex, exp-concave and convex loss functions.
  arXiv  Detail & Related papers  (2023-07-17T09:55:35Z)
• Cross-Entropy Loss Functions: Theoretical Analysis and Applications [27.3569897539488]
  We present a theoretical analysis of a broad family of loss functions, that includes cross-entropy (or logistic loss), generalized cross-entropy, the mean absolute error and other cross-entropy-like loss functions. We show that these loss functions are beneficial in the adversarial setting by proving that they admit $H$-consistency bounds. This leads to new adversarial robustness algorithms that consist of minimizing a regularized smooth adversarial comp-sum loss.
  arXiv  Detail & Related papers  (2023-04-14T17:58:23Z)
• FeDXL: Provable Federated Learning for Deep X-Risk Optimization [105.17383135458897]
  We tackle a novel federated learning (FL) problem for optimizing a family of X-risks, to which no existing algorithms are applicable. The challenges for designing an FL algorithm for X-risks lie in the non-decomability of the objective over multiple machines and the interdependency between different machines.
  arXiv  Detail & Related papers  (2022-10-26T00:23:36Z)
• Horizon-Free and Variance-Dependent Reinforcement Learning for Latent Markov Decision Processes [62.90204655228324]
  We study regret minimization for reinforcement learning (RL) in Latent Markov Decision Processes (LMDPs) with context in hindsight. We design a novel model-based algorithmic framework which can be instantiated with both a model-optimistic and a value-optimistic solver.
  arXiv  Detail & Related papers  (2022-10-20T21:32:01Z)
• Provably Efficient Offline Reinforcement Learning with Trajectory-Wise Reward [66.81579829897392]
  We propose a novel offline reinforcement learning algorithm called Pessimistic vAlue iteRaTion with rEward Decomposition (PARTED) PARTED decomposes the trajectory return into per-step proxy rewards via least-squares-based reward redistribution, and then performs pessimistic value based on the learned proxy reward. To the best of our knowledge, PARTED is the first offline RL algorithm that is provably efficient in general MDP with trajectory-wise reward.
  arXiv  Detail & Related papers  (2022-06-13T19:11:22Z)
• Realizing GANs via a Tunable Loss Function [7.455546102930911]
  We introduce a tunable GAN, called $alpha$-GAN, parameterized by $alpha in (0,infty]$. We show that $alpha$-GAN is intimately related to the Arimoto divergence.
  arXiv  Detail & Related papers  (2021-06-09T17:18:21Z)
• A surrogate loss function for optimization of $F_\beta$ score in binary classification with imbalanced data [0.0]
  The gradient paths of the proposed surrogate $F_beta$ loss function approximate the gradient paths of the large sample limit of the $F_beta$ score. It is demonstrated that the proposed surrogate $F_beta$ loss function is effective for optimizing $F_beta$ scores under class imbalances.
  arXiv  Detail & Related papers  (2021-04-03T18:36:23Z)
• Piecewise Linear Regression via a Difference of Convex Functions [50.89452535187813]
  We present a new piecewise linear regression methodology that utilizes fitting a difference of convex functions (DC functions) to the data. We empirically validate the method, showing it to be practically implementable, and to have comparable performance to existing regression/classification methods on real-world datasets.
  arXiv  Detail & Related papers  (2020-07-05T18:58:47Z)
• Supervised Learning: No Loss No Cry [51.07683542418145]
  Supervised learning requires the specification of a loss function to minimise. This paper revisits the sc SLIsotron algorithm of Kakade et al. (2011) through a novel lens. We show how it provides a principled procedure for learning the loss.
  arXiv  Detail & Related papers  (2020-02-10T05:30:52Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.