Tag
该论文研究了不同训练策略与损失函数对长尾/不平衡数据上集值分类器（SVC）性能的影响，并改进了 odd-KK-out（OKO）损失，将其计算开销从3.2×降至1.36×，显著提升了多个SVC方法的宏平均准确率。
This paper challenges the assumption that EER is a reliable proxy for human voice similarity in speech generation, demonstrating that perceptual alignment depends more on training objectives than model performance. It proposes embedding geometry with effective dimensionality as a principled metric for evaluating voice similarity.
The tweet highlights the value of understanding different AI fine-tuning methods like SFT and RL, their impact on model behavior, and recommends using mathematical analysis and visual aids such as ASCII art to grasp these concepts.
The paper examines XGBoost's sensitivity to outliers in regression tasks and introduces robust loss functions based on M-, S-, and τ-estimators, proposing MM-XGBoost for a better trade-off between robustness and predictive accuracy.
Discussion of loss functions in instance representation learning, focusing on the use of NCE to approximate the computationally infeasible MLE objective.
This paper formalizes the concept of Bayes-sufficient representations in supervised learning, defining when a representation retains exactly the information needed for Bayes-optimal prediction under a given loss function. It introduces the Bayes quotient as a canonical loss-dependent object and connects the framework to property elicitation, illustrating distinctions between sufficiency, minimality, and excess retained information through experiments.
This paper studies symmetrization of loss functions for robust training under label noise, introducing SGCE and alpha-MAE loss functions that interpolate between multi-class unhinged loss and Mean Absolute Error, with theoretical guarantees and competitive empirical performance.
This paper identifies a critical 'model collapse' issue in standard fine-tuning for causal reasoning and proposes a semantic loss function with graph-based logical constraints to prevent it.
This paper proposes topological optimal transport-based loss functions for improving structured recipe generation in language models, addressing the limitations of standard cross-entropy training by better handling ingredient composition, quantities, and procedural accuracy. The approach shows significant improvements on recipe-specific metrics with 62% human preference over baseline methods.