Qisen Cheng

In this paper, we propose ShapTST, a framework that enables time-series transformers to efficiently generate Shapley-value-based explanations alongside predictions in a single forward pass. Shapley values are widely used to evaluate the contribution of different time-steps and features in a test sample, and are commonly generated through repeatedly inferring on each sample with different parts of information removed. Therefore, it requires expensive inference-time computations that occur at every request for model explanations. In contrast, our framework unifies the explanation and prediction in training through a novel Shapley-based pre-training design, which eliminates the undesirable test-time computation and replaces it with a single-time pre-training. Moreover, this specialized pre-training benefits the prediction performance by making the transformer model more effectively weigh different features and time-steps in the time-series, particularly improving the robustness against data noise that is common to raw time-series data. We experimentally validated our approach on eight public datasets, where our time-series model achieved competitive results in both classification and regression tasks, while providing Shapley-based explanations similar to those obtained with post-hoc computation. Our work offers an efficient and explainable solution for time-series analysis tasks in the safety-critical applications.

Fuzzy Graph Attention Network (FGAT), which combines Fuzzy Rough Sets and Graph Attention Networks, has shown promise in tasks requiring robust graph-based learning. However, existing models struggle to effectively capture dependencies from multiple perspectives, limiting their ability to model complex data. To address this gap, we propose the Multi-view Fuzzy Graph Attention Network (MFGAT), a novel framework that constructs and aggregates multi-view information using a specially designed Transformation Block. This block dynamically transforms data from multiple aspects and aggregates the resulting representations via a weighted sum mechanism, enabling comprehensive multi-view modeling. The aggregated information is fed into FGAT to enhance fuzzy graph convolutions. Additionally, we introduce a simple yet effective learnable global pooling mechanism for improved graph-level understanding. Extensive experiments on graph classification tasks demonstrate that MFGAT outperforms state-of-the-art baselines, underscoring its effectiveness and versatility.

This paper reports on a bidirectional Knudsen pump (KP) with a 3D-printed thermal management platform; the pump is intended principally for microscale gas chromatography applications. Knudsen pumps utilize thermal transpiration, where non-viscous flow is created against a temperature gradient; no moving parts are necessary. Here, a specialized design leverages 3D direct metal laser sintering and provides thermal management that minimizes loss from a joule heater located on the outlet side of KP, while maintaining convective cooling on the inlet side. The 3D-KP design is integrative and compact, and is specifically intended to simplify assembly. The 3D-KP pumping area is ≈1.1 cm²; with the integrated heat sink, the structure has a footprint of 64.2 × 64.2 mm². Using mixed cellulose ester (MCE) membranes with a 25 nm average pore diameter and 525 μm total membrane thickness as the pumping media, the 3D-KP achieves a maximum flow rate of 0.39 sccm and blocking pressure of 818.2 Pa at 2 W input power. The operating temperature is 72.2 °C at ambient room temperature. In addition to MCE membranes, anodic aluminum oxide (AAO) membranes are evaluated as the pumping media; these AAO membranes can accommodate higher operating temperatures than MCE membranes. The 3D-KP with AAO membranes with 0.2 μm average pore diameter and 531 μm total membrane thickness achieves a maximum flow rate of 0.75 sccm and blocking pressure of 496.1 Pa at 9.8 W at an operating temperature of 191.2 °C.