Method for multi-task learning fusion network traffic classification to address small sample labels.

In the context of the proliferated evolution of network service types and the expeditious augmentation of network resource deployment, the requisition for copious labeled datasets to facilitate superior performance in traffic classification methods, particularly those hinging on deep learning, is imperative. Nonetheless, the procurement and annotation of such extensive datasets necessitate considerable temporal and human resource investments. In response to this predicament, this work introduces a methodology, termed MTEFU, leveraging a deep learning model-based multi-task learning algorithm, strategically designed to mitigate the reliance on substantial labeled training samples. Multiple classification tasks, encompassing duration, bandwidth size, and business traffic category, are incorporated, with a shared parameter strategy implemented amongst tasks to assure the transference of information across disparate tasks. Employing CNN, SAE, GRU, and LSTM as multi-task learning classification models, training validation and experimental testing were conducted on the QUIC dataset. A comparative analysis with single-task and ensemble learning methods reveals that, in the context of predicting network traffic types, the accuracy derived from the multi-task learning strategy, even with a mere 150 labeled samples, can emulate the 94.67% accuracy achieved through single-task learning with a fully labeled dataset of 6139 samples.

Molecular basis for the catalytic mechanism of human neutral sphingomyelinases 1 (hSMPD2).

Enzymatic breakdown of sphingomyelin by sphingomyelinase (SMase) is the main source of the membrane lipids, ceramides, which are involved in many cellular physiological processes. However, the full-length structure of human neutral SMase has not been resolved; therefore, its catalytic mechanism remains unknown. Here, we resolve the structure of human full-length neutral SMase, sphingomyelinase 1 (SMPD2), which reveals that C-terminal transmembrane helices contribute to dimeric architecture of hSMPD2 and that D111 - K116 loop domain is essential for substrate hydrolysis. Coupled with molecular docking, we clarify the binding pose of sphingomyelin, and site-directed mutagenesis further confirms key residues responsible for sphingomyelin binding. Hybrid quantum mechanics/molecular mechanics (QM/MM) molecular dynamic (MD) simulations are utilized to elaborate the catalysis of hSMPD2 with the reported in vitro substrates, sphingomyelin and lyso-platelet activating fator (lyso-PAF). Our study provides mechanistic details that enhance our knowledge of lipid metabolism and may lead to an improved understanding of ceramide in disease and in cancer treatment.

Geometry-Gradient Magnetocontrollable Lubricant-Infused Microwall Array for Passive/Active Hybrid Bidirectional Droplet Transport.

Manipulation of droplets has increasingly garnered global attention, owing to its multifarious potential applications, including microfluidics and medical diagnostic tests. To control the droplet motion, geometry-gradient-based passive transport has emerged as a well-established strategy, which induces a Laplace pressure difference based on the droplet radius differences in confined state and transport droplets with no consumption of external energy, whereas this transportation method has inevitably shown some critical limitations: unidirectionality, uncontrollability, short moving distance, and low velocity. Herein, a magnetocontrollable lubricant-infused microwall array (MLIMA) is designed as a key solution to this issue. In the absence of a magnetic field, droplets can spontaneously travel from the tip toward the root of the structure as a result of the geometry-gradient-induced Laplace pressure difference. When the subject of an external magnetic field, the microwalls bend and overlap sequentially, ultimately resulting in the formation of a continuous slippery meniscus surface. The formed meniscus surface can exert sufficient propulsive force to surmount the Laplace pressure difference of the droplet, thereby effectuating active transport. Through the continuous movement of the microwalls, droplets can be actively transported against the Laplace pressure difference from the root to the tip side of the MLIMA or continue to actively move to the root after finishing the passive self-transport. This work demonstrates passive/active hybrid bidirectional droplet transport capabilities, validates its feasibility in the accurate control of droplet manipulation, and exhibits great potential in chemical microreactions, bioassays, and the medical field.