Xiubin Fan | Searcharxiv

arXiv2026

GenHAR: Generalizing Cross-domain Human Activity Recognition for Last-mile Delivery

Human Activity Recognition (HAR) has shown remarkable effectiveness in various applications, such as smart healthcare and intelligent manufacturing. However, a major challenge faced by HAR is the distribution shift across different sensor data domains, which often leads to decreased performance when deployed for real-world applications. To address this issue, this paper introduces GenHAR, a novel framework designed to mitigate the domain gap by learning domain-invariant sensor representations. GenHAR aims to enhance the generalization capabilities of HAR on target domains purely with data from the source domain. The key novelty of GenHAR lies in two aspects. Firstly, GenHAR tokenizes sensor data and learns correlations among frequency sensor channel dimensions to improve the robustness of HAR models. Secondly, GenHAR improves the efficiency via selective masking and an efficient attention mechanism. We conduct a systematic analysis of GenHAR by comparing it with state-of-the-art HAR methods on real-world human activity datasets. Results show that GenHAR outperforms state-of-the-art methods by 9.97% in accuracy, and reduces Floating Point Operations by 6.4 times. Moreover, we deploy GenHAR at a leading logistics company in 4 cities, and have detected 2.15 billion real-time activities. We release our code at: https://github.com/Sensor-FoundationModel/GenHAR.

cs.CV↗

arXiv2014

Asymptotic Granularity Reduction and Its Application

It is well known that the inverse function of y = x with the derivative y' = 1 is x = y, the inverse function of y = c with the derivative y' = 0 is inexistent, and so on. Hence, on the assumption that the noninvertibility of the univariate increasing function y = f(x) with x > 0 is in direct proportion to the growth rate reflected by its derivative, the authors put forward a method of comparing difficulties in inverting two functions on a continuous or discrete interval called asymptotic granularity reduction (AGR) which integrates asymptotic analysis with logarithmic granularities, and is an extension and a complement to polynomial time (Turing) reduction (PTR). Prove by AGR that inverting y = x ^ x (mod p) is computationally harder than inverting y = g ^ x (mod p), and inverting y = g ^ (x ^ n) (mod p) is computationally equivalent to inverting y = g ^ x (mod p), which are compatible with the results from PTR. Besides, apply AGR to the comparison of inverting y = x ^ n (mod p) with y = g ^ x (mod p), y = g ^ (g1 ^ x) (mod p) with y = g ^ x (mod p), and y = x ^ n + x + 1 (mod p) with y = x ^ n (mod p) in difficulty, and observe that the results are consistent with existing facts, which further illustrates that AGR is suitable for comparison of inversion problems in difficulty. Last, prove by AGR that inverting y = (x ^ n)(g ^ x) (mod p) is computationally equivalent to inverting y = g ^ x (mod p) when PTR can not be utilized expediently. AGR with the assumption partitions the complexities of problems more detailedly, and finds out some new evidence for the security of cryptosystems.

cs.CC↗