Vector quantization-based image semantic communication systems have successfully boosted transmission efficiency. However, existing models like Vector Quantized Variational Autoencoder 2 (VQVAE2) suffer from slow training convergence and inefficient codebook utilization. This paper proposes a Dynamic Codebook Update (DCU) mechanism for vector quantization models to address these issues during training. By monitoring the usage frequency of embedding vectors and adding noise to inactive ones, the DCU mechanism dynamically adjusts these vectors to enhance codebook utilization. Experiments demonstrate that the DCU model achieves faster convergence and superior image reconstruction quality compared to the traditional VQVAE2 model.

Accurately predicting user watch-time is crucial for enhancing user stickiness and retention in video recommendation systems. Existing watch-time prediction approaches typically involve transformations of watch-time labels for prediction and subsequent reversal, ignoring both the natural distribution properties of label and the instance representation confusion that results in inaccurate predictions. In this paper, we propose ProWTP, a two-stage method combining prototype learning and optimal transport for watch-time regression prediction, suitable for any deep recommendation model. Specifically, we observe that the watch-ratio (the ratio of watch-time to video duration) within the same duration bucket exhibits a multimodal distribution. To facilitate incorporation into models, we use a hierarchical vector quantised variational autoencoder (HVQ-VAE) to convert the continuous label distribution into a high-dimensional discrete distribution, serving as credible prototypes for calibrations. Based on this, ProWTP views the alignment between prototypes and instance representations as a Semi-relaxed Unbalanced Optimal Transport (SUOT) problem, where the marginal constraints of prototypes are relaxed. And the corresponding optimization problem is reformulated as a weighted Lasso problem for solution. Moreover, ProWTP introduces the assignment and compactness losses to encourage instances to cluster closely around their respective prototypes, thereby enhancing the prototype-level distinguishability. Finally, we conducted extensive offline experiments on two industrial datasets, demonstrating our consistent superiority in real-world application.

In video recommendation systems, accurately predicting watch time is crucial for enhancing user engagement and retention. Traditional methods typically apply label transformations or mitigate duration bias to improve performance but overlook that erroneous instance representations are the primary cause of significant prediction errors. Moreover, these approaches predominantly rely on point perdition, limiting their robustness. To address these challenges, we propose LDA, a novel prediction paradigm that optimizes instance representations by explicitly leveraging label distributions as anchors within the model, enabling more accurate and robust predictions. Our analysis reveals that watch ratio across different duration groups exhibit distinct multi-peak distributions, reflecting the strong aggregation of user behavior. Based on this finding, we employ Vector Quantized Variational Auto-encoder (VQ-VAE) to convert the continuous watch ratio distribution into representative anchors that capture these multi-peak characteristics within each duration group. Subsequently, we project both instance representations and anchors into a common space and utilize Optimal Transport (OT) to generate pseudo-labels aligned with the anchor distribution, allowing instances to obtain structured coordinates within this space during training. Finally, we derive optimized instance representations for watch time prediction by aggregating anchor vectors through weighted integration. Extensive offline experiments on two datasets and large-scale online A/B testing on a short-video platform with over 300 million DAUs demonstrate the consistent superiority of LDA in watch time prediction.

Pretraining molecular representations is crucial for drug and material discovery. Recent methods focus on learning representations from geometric structures, effectively capturing 3D position information. Yet, they overlook the rich information in biomedical texts, which detail molecules’ properties and substructures. With this in mind, we set up a data collection effort for 200K pairs of ground-state geometric structures and biomedical texts, resulting in a PubChem3D dataset. Based on this dataset, we propose the GeomCLIP framework to enhance geometric pretraining and understanding by biomedical texts. During pre-training, we design two types of tasks, i.e., multimodal representation alignment and unimodal denoising pretraining, to align the 3D geometric encoder with textual information and, at the same time, preserve its original representation power. Experimental results show the effectiveness of GeomCLIP in various tasks such as molecule property prediction, zero-shot text-molecule retrieval, and 3D molecule captioning. Our code and collected dataset are available at https://github.com/xiaocui3737/GeomCLIP.

This paper introduces a novel generalized selfimitation learning (GSIL) framework, which effectively and efficiently aligns large language models with offline demonstration data.We develop GSIL by deriving a surrogate objective of imitation learning with density ratio estimates, facilitating the use of self-generated data and optimizing the imitation learning objective with simple classification losses.GSIL eliminates the need for complex adversarial training in standard imitation learning, achieving lightweight and efficient fine-tuning for large language models.In addition, GSIL encompasses a family of offline losses parameterized by a general class of convex functions for density ratio estimation and enables a unified view for alignment with demonstration data.Extensive experiments show that GSIL consistently and significantly outperforms baselines in many challenging benchmarks, such as coding (HuamnEval), mathematical reasoning (GSM8K) and instruction-following benchmark (MT-Bench).Code is public available at https://github.com/tengxiao1/GSIL.

Heavy hitter detection is a fundamental network measurement task that provides critical support for many network applications. However, achieving flexible and fast heavy hitter detection in massive network traffic is challenging. Existing works generally perform detection by tracking large flows that may become heavy hitters, but they struggle to accurately identify these large flows, leading to poor accuracy. In this paper, we propose an efficient detection algorithm called HeavyCuckoo, which shows high flexibility and fast processing. We track only those large flows likely to be heavy hitters, replacing small flows of limited use for detection by exploring the activity of flow arrivals. During replacement, we utilize a tailored Conservative Replacement strategy and a tailored Selective Cuckoo Hash strategy to avoid large flows from being replaced incorrectly. We conduct theoretical analyses of memory space complexity and time complexity, and provide the error bound for heavy hitter detection. Our proposed algorithm is evaluated on real-world Internet traffic traces. Experimental results show that, compared to the prior art, the algorithm improves the Fβ-score by 56.94% and achieves 1.4724 times throughput.

Numerous hyperspectral image (HSI) super-resolution (SR) approaches have been proposed and attained remarkable performance recently. However, the enormous computational and memory costs of the deeper and heavier networks make the application challenging in the actual scene. Therefore, a lightweight but efficient local-global feature fusion module network (LGFFMN) is proposed to reduce the computational burden while achieving superior reconstruction performance. In particular, a module LFFM is proposed to utilize CNN for effective local feature extraction. Meanwhile, to exploit global information and avoid the computational burden of traditional Transformers, a multi-scale representation-based feature modulation mechanism is adopted to build a Vision Transformer (ViT)-like block GFFM. In order to reduce the high dimension of HSI while utilizing the higher similarity among adjacent bands, the overall processing manner is group by group. Experiments demonstrate that our LGFFMN achieves a satisfactory balance in SR reconstruction performance and efficiency metrics and is highly competitive compared with other state-of-the-art HSI SR methods.

Modular parallel inverter technology can enhance the power level and redundancy of wireless power transfer (WPT) systems, contributing to standardized production. It serves as an effective method for realizing high-power systems. However, inappropriate selection of compensation components can negatively affect the system’s efficiency, power factor, and the flexibility of modularization. This paper analyzes two key properties of modular-parallel-inverter WPT (MPI-WPT) systems: module number flexibility and modular deviation suppression. Firstly, to assess the modular deviation suppression of the system, its definition and calculation method are provided. Secondly, the expansion condition to achieve modular flexibility is examined, highlighting that the modular system needs to be a fully resonant system. Subsequently, an expansion design methodology from a single WPT system to an MPI-WPT system is proposed, ensuring the preservation of the original properties of the system during its extension. Finally, the parallel characteristics of MPI-WPT systems were experimentally verified.

This article presents an 8 × 8 lattice-reduction-aided (LRA) soft-output multiple-input multiple-output (MIMO) detector for Chinese enhanced ultrahigh throughput (EUHT) wireless local area network (LAN) standard. The preprocessing algorithm combining simplified-sorting Cholesky decomposition and low-complexity decoupled lattice reduction (LDLR) is proposed to reduce computational complexity and latency with parallelism improvement. In addition, K-best detection adopts a sorting-reduced strategy utilizing approximate ordered sequence. Compared with other published LRA K-best detection algorithms, simulation results show that our proposed algorithm has performance improvement. In addition, in order to save hardware resources, a folded K-best architecture and an optimized intermediate storage strategy are introduced. Furthermore, a fully pipelined VLSI architecture is designed in Semiconductor Manufacturing International Corporation (SMIC) 40-nm 1P9M technology to support the 8 × 8.64 -QAM MIMO-OFDM system. The detector can achieve 3.85-Gb/s data throughput at 641-MHz clock frequency with 0.71-μs latency. The proposed detector is competitive in terms of latency, throughput, and area efficiency to state-of-the-art works and can meet the data-rate requirement of the EUHT standard.

Inductive Power Transfer (IPT) technology has received extensive application research due to its safety and convenience. As a circuit topology with high efficiency and high controllability, dual active bridge (DAB) has been used in IPT system in recent years. In high-power applications, the EMI generated by the power device switch becomes more significant, which in turn negatively affects the efficiency and even the stability of the IPT system. There are several ways to reduce EMI. This article studies from the perspective of zero voltage switching (ZVS) of the DAB. When the duty cycle of the full bridge changes, the soft switching may not be maintained. This paper calculates and analyzes the ZVS operating range of the system under different situations. On this basis, a control method is proposed to achieve ZVS in the full power range of the system by adjusting the bilateral load angle. And the efficiency and stability of the system are improved. The correctness of the above theory and control method is verified through simulations.

With development in wide band-gap semiconductors, such as silicon carbide (SiC), inductive power transfer (IPT) has become a promising technology for charging electric vehicles (EV). To meet fast charging demands by consumers, higher power levels in IPT is required. The power in IPT is usually limited by thermal stress due to losses at the vehicle side. Synchronous rectification is an efficient way to reduce losses compared to passive rectification. In this paper, the loss reduction accounted to synchronous rectification is quantitatively evaluated. The level of reductions varies depending on the load level, primary voltage, and constant current or constant voltage (CCCV) operation. The analytical results support the simulation results for various operating points. The loss profiles are studied for different dc-link voltages and duty cycles. An 800 V, 50 kW dual-active bridge test setup with series-series compensation is constructed and experiments are done for verification. The setup is arranged in a back-to-back configuration with a common dc-link. Experimental result shows that losses are reduced by up to 60 % on the receiving side with synchronous rectification. At rated operation the losses are reduced by 100 W in the secondary side inverter.