Ensuring algorithmic fairness is a central challenge in high-stakes machine learning applications, where biased predictions can have harmful societal consequences. Discrimination may persist even when sensitive features are excluded from model inputs, due to statistical dependencies between protected and ostensibly non-sensitive attributes that allow bias to be indirectly learned. Mutual information is a principled measure of such dependence, yet its non-differentiability under hard-threshold classification rules precludes direct use in gradient-based training. This paper proposes a fairness-aware learning strategy that embeds a differentiable surrogate of mutual information directly into the objective of nonlinear classifiers. The surrogate replaces hard label assignment with a Bernoulli relaxation of predicted probabilities, stabilizes empirical distributions via kernel smoothing, and accounts for epistemic uncertainty through Monte Carlo dropout, yielding an uncertainty-aware regularizer compatible with stochastic gradient optimization. Experiments on synthetic and real-world datasets show that the method achieves favorable fairness–accuracy trade-offs, with substantial bias reduction and minimal impact on predictive performance. Additional evaluations confirm robustness under counterfactual perturbations and reveal sublinear runtime growth with dataset size, contrasting favorably with the superlinear scaling of a genetic algorithm baseline. Overall, the framework offers a general, theoretically grounded, and practically viable approach to fairness-aware learning in complex, high-dimensional settings, supporting responsible algorithmic decision-making alongside expressive modeling.

MOTIVATION: Data visualization is increasingly important in genomics, enabling researchers to uncover inheritance and recombination patterns across generations. While most existing tools focus on ancestry prediction, they lack functionality for analyzing known ancestries in controlled settings, such as determining parental contributions to offspring genomes. To address this gap, I developed pepa, a lightweight, deterministic, modular tool that visualizes and quantifies genomic inheritance, designed for beginner and advanced users. RESULTS: pepa is a program for processing VCF files, assigning ancestries to homozygous SNPs, and clustering them into biologically meaningful regions. It generates human-readable comparison tables and visualizes inheritance patterns with chromosome paintings through R. Tested on fission yeast, pepa revealed non-uniform recombination patterns, with chromosomes largely inherited from one parent and seemingly random recombination. Quantitative analyses showed differences in parental contributions at the nucleotide and gene levels, with some offspring inheriting similar percentages from parents. However, the painted chromosomes revealed that even offspring with similar percentages from one parent rarely inherit the same genomic region, highlighting the importance of this tool in drawing biologically meaningful insights. pepa provides an accessible and powerful solution for analyzing genomic inheritance, bridging experimental and computational biology. Its modular design and minimal dependencies allow adaptation to diverse organisms, facilitating intuitive visualization and quantitative insights into recombination dynamics.

Abstract Knowledge distillation is recognized as a valuable model compression strategy that alleviates the computational burden of large language models while preserving performance. This strategy involves training a smaller model utilizing both real data and predictions from a more cumbersome model. Traditional distillation methods, however, are often compromised by exposure bias, which results from reliance on next-step prediction training loss. This bias emerges when models are tested in free-running mode, differing from their training regime and leading to a progressive drift in input distributions between testing and training phases. An analogous issue, known as ‘distributional shift’, has been effectively addressed in imitation learning through various methodologies. Therefore, this paper specifically tailors an imitation learning-based solution to a traditional knowledge distillation framework which inherently considers both real data and the teacher’s predictions as dual sources of expert demonstrations. The effectiveness of this approach is demonstrated over five different test datasets, where it outperforms traditional benchmarks across all evaluation metrics. Specifically, it achieves superior results in perplexity, multi-token generation, and G-Eval score, indicating improvements in both predictive accuracy and alignment with human judgment in text quality. These results underscore the potential of this approach to effectively address exposure bias in large language model distillation.