Diffusion-Driven Two-Stage Active Learning for Low-Budget Semantic Segmentation

Contemporary machine learning models, particularly large language models, excel in static tasks but struggle in dynamic environments due to inflexible architectures. This research introduces dynamic nested hierarchies, allowing models to autonomously adapt their optimization levels and structures during training or inference. Inspired by neuroplasticity, this approach aims to overcome limitations in lifelong learning by enabling continuous adaptation without predefined constraints.

The Gini score is a widely used metric in statistical modeling and machine learning for validating and selecting models. Traditionally applied in binary contexts, it has equivalent formulations like the receiver operating characteristic (ROC) and area under the curve (AUC). This paper extends the Gini score to accommodate ties in risk rankings and adapts it for scenarios involving case weights, which is common in actuarial literature.

This study introduces a novel model-agnostic attribution method for time series classification, focusing on causal attributions rather than traditional associational methods. The approach assesses the causal effects of predefined segments within time series on classification outcomes, utilizing advanced diffusion models and state space models for estimating counterfactual outcomes. The findings highlight the insights gained from this method across various time series classification tasks.

The article discusses the challenges faced in machine learning as applications become more complex and require more than just accuracy. It highlights the prevalent method of aggregating penalties for requirement violations into training objectives, which necessitates careful tuning of hyperparameters. This tuning process can become ineffective with a moderate number of requirements, especially when dealing with equality constraints related to fairness. The work presented derives a generalization theory for equality-constrained statistical learning problems.

This technical note discusses the derivative of the truncated singular value and eigenvalue decomposition, which is crucial for applications in machine learning and computational physics. It emphasizes the need for stable and efficient linear algebra gradient computations, particularly in the context of automatic differentiation techniques. The note builds on previous work, providing a detailed explanation of how to derive the relevant terms while focusing on expressing the derivative concerning the truncated part of the decomposition.

A new machine learning-based multimodal framework has been developed for wearable sensor-based archery action recognition and stress estimation. This innovative system utilizes a wrist-worn device equipped with an accelerometer and photoplethysmography (PPG) sensor to collect synchronized motion and physiological data during archery sessions. The framework achieves high accuracy in motion recognition and stress estimation, marking a significant advancement in the analysis of athletes' performance in precision sports.

The article discusses the concept of soft-label training in machine learning, which preserves epistemic uncertainty by treating annotation distributions as ground truth. Traditional methods often collapse diverse human judgments into single labels, leading to misalignment between model certainty and human perception. Empirical results show that soft-label training reduces KL divergence from human annotations by 32% and enhances correlation between model and annotation entropy by 61%, while maintaining accuracy comparable to hard-label training.

Mobile jammers present a significant threat to 5G networks, especially in military settings. An innovative anti-jamming framework has been proposed, utilizing Multiple Signal Classification (MUSIC) for precise Direction-of-Arrival (DoA) estimation and Minimum Variance Distortionless Response (MVDR) beamforming for adaptive interference suppression. Machine learning enhances DoA prediction for mobile jammers. Simulations indicate an average Signal-to-Noise Ratio (SNR) improvement of 9.58 dB and a DoA estimation accuracy of up to 99.8%, showcasing the framework's effectiveness in dynamic environ…