Jiachen Xiao | Searcharxiv

arXiv2026

Talking Ink: A Flow-Based Multi-Molecule Molecular Communication Testbed for Effective Channel Modeling and Detector Benchmarking

Practical molecular communication (MC) testbeds are essential for translating theoretical concepts toward future applications while retaining physical realism, accessibility, and experimental repeatability. We present a flowbased MC experimental system for real-time multi-molecule (MUMO) transmission using cyan, magenta, and yellow inks as distinguishable molecule-like signaling carriers. Three micropumps and a multi-needle injection system release the inks into a background-flow channel, and a non-invasive spectral sensor estimates received color traces at variable transmitter-receiver distances. The platform supports measurements of isolated-pulse responses across distances. We interpret these measurements as effective end-to-end channel impulse responses (CIRs) shaped by finite release, flow propagation, and receiver readout, and compare two compact model families. Both reproduce much of the dominant arrival timing and pulse shape, while late-tail mismatch remains the main limitation. We connect this channel characterization to communication performance through continuous MUMO on-off keying (OOK) payload measurements. The resulting retrospective benchmark with full-data parameter optimization includes traceonly and CIR-assisted methods. We introduce modified energydifference detection (MEDD), an adaptive extension of standard energy-difference detection (EDD). MEDD outperforms all other trace-only detectors with 34/4500 = 0.76% bit error rate (BER) and closely approaches the best overall minimum mean-square error (MMSE) result of 32/4500 = 0.71%. After targeted parameter tuning, the detector benchmark produces zero errors over the evaluated payloads at 6 bit/s over 8 cm and at 3 bit/s over 24 cm. Overall, the results establish the platform as a reusable experimental setup for effective channel modeling, practical detector benchmarking, and future multi-molecule networking experiments.

cs.ET↗

arXiv2024

A Pure Transformer Pretraining Framework on Text-attributed Graphs

Pretraining plays a pivotal role in acquiring generalized knowledge from large-scale data, achieving remarkable successes as evidenced by large models in CV and NLP. However, progress in the graph domain remains limited due to fundamental challenges such as feature heterogeneity and structural heterogeneity. Recently, increasing efforts have been made to enhance node feature quality with Large Language Models (LLMs) on text-attributed graphs (TAGs), demonstrating superiority to traditional bag-of-words or word2vec techniques. These high-quality node features reduce the previously critical role of graph structure, resulting in a modest performance gap between Graph Neural Networks (GNNs) and structure-agnostic Multi-Layer Perceptrons (MLPs). Motivated by this, we introduce a feature-centric pretraining perspective by treating graph structure as a prior and leveraging the rich, unified feature space to learn refined interaction patterns that generalizes across graphs. Our framework, Graph Sequence Pretraining with Transformer (GSPT), samples node contexts through random walks and employs masked feature reconstruction to capture pairwise proximity in the LLM-unified feature space using a standard Transformer. By utilizing unified text representations rather than varying structures, our framework achieves significantly better transferability among graphs within the same domain. GSPT can be easily adapted to both node classification and link prediction, demonstrating promising empirical success on various datasets.

cs.AI↗