Search

Mixture of Experts (MoE) offers remarkable performance and computational efficiency by selectively activating subsets of model parameters. Traditionally, MoE models use homogeneous experts, each with identical capacity. However, varying complexity in input data necessitates experts with diverse capabilities, while homogeneous MoE hinders effective expert specialization and efficient parameter utilization. In this study, we propose a novel Heterogeneous Mixture of Experts (HMoE), where experts differ in size and thus possess diverse capacities. This heterogeneity allows for more specialized experts to handle varying token complexities more effectively. To address the imbalance in expert activation, we propose a novel training objective that encourages the frequent activation of smaller experts, enhancing computational efficiency and parameter utilization. Extensive experiments demonstrate that HMoE achieves lower loss with fewer activated parameters and outperforms conventional homogeneous MoE models on various pre-training evaluation benchmarks. Codes will be released upon acceptance.

Mid-infrared spectrum is a critical tool for chemical analysis, industrial inspection, environment, and other fields due to its rich chemical bond information. However, the complicated growth or fabrication procedures of existing mid-infrared sensitive materials hinder the large-scale production and utilization of mid-infrared detectors. To address this issue, we developed Si:S barrier detectors employing sulfur doped silicon and a sophisticated band barrier design. Since the transport of dark current and photo current is separated, the barrier design effectively suppresses the dark current while allowing the photo current to leverage gain mechanisms, thereby substantially improving signal-to-noise ratio. As a result, the detector exhibits an infrared response range covering from 1.12 to 4.4{\mu}m with a peak at 3.3{\mu}m, excluding its intrinsic response in visible range. Its peak quantum efficiency surpasses that of the best mid-infrared silicon-based detector reported to date by an order of magnitude, reaching 2% at room temperature. The peak detectivity at 90K is 1.4E11 Jones @1.4V and decreases to 4.4E9 Jones @1.4V, 210K, comparable to the typical III-V and IV-VI photodetectors at one thousandth fabrication cost. Leveraging the well-established silicon-based manufacturing process, this device holds promise for large-scale production at a reduced price, offering a cost-effective solution for future mid-infrared detection.

Large vision-language models (LVLMs) are markedly proficient in deriving visual representations guided by natural language. Recent explorations have utilized LVLMs to tackle zero-shot visual anomaly detection (VAD) challenges by pairing images with textual descriptions indicative of normal and abnormal conditions, referred to as anomaly prompts. However, existing approaches depend on static anomaly prompts that are prone to cross-semantic ambiguity, and prioritize global image-level representations over crucial local pixel-level image-to-text alignment that is necessary for accurate anomaly localization. In this paper, we present ALFA, a training-free approach designed to address these challenges via a unified model. We propose a run-time prompt adaptation strategy, which first generates informative anomaly prompts to leverage the capabilities of a large language model (LLM). This strategy is enhanced by a contextual scoring mechanism for per-image anomaly prompt adaptation and cross-semantic ambiguity mitigation. We further introduce a novel fine-grained aligner to fuse local pixel-level semantics for precise anomaly localization, by projecting the image-text alignment from global to local semantic spaces. Extensive evaluations on the challenging MVTec and VisA datasets confirm ALFA's effectiveness in harnessing the language potential for zero-shot VAD, achieving significant PRO improvements of 12.1% on MVTec AD and 8.9% on VisA compared to state-of-the-art zero-shot VAD approaches.

Weakly supervised text classification (WSTC), also called zero-shot or dataless text classification, has attracted increasing attention due to its applicability in classifying a mass of texts within the dynamic and open Web environment, since it requires only a limited set of seed words (label names) for each category instead of labeled data. With the help of recently popular prompting Pre-trained Language Models (PLMs), many studies leveraged manually crafted and/or automatically identified verbalizers to estimate the likelihood of categories, but they failed to differentiate the effects of these category-indicative words, let alone capture their correlations and realize adaptive adjustments according to the unlabeled corpus. In this paper, in order to let the PLM effectively understand each category, we at first propose a novel form of rule-based knowledge using logical expressions to characterize the meanings of categories. Then, we develop a prompting PLM-based approach named RulePrompt for the WSTC task, consisting of a rule mining module and a rule-enhanced pseudo label generation module, plus a self-supervised fine-tuning module to make the PLM align with this task. Within this framework, the inaccurate pseudo labels assigned to texts and the imprecise logical rules associated with categories mutually enhance each other in an alternative manner. That establishes a self-iterative closed loop of knowledge (rule) acquisition and utilization, with seed words serving as the starting point. Extensive experiments validate the effectiveness and robustness of our approach, which markedly outperforms state-of-the-art weakly supervised methods. What is more, our approach yields interpretable category rules, proving its advantage in disambiguating easily-confused categories., Comment: Accepted by WWW 2024

Graph Attention Network (GAT) is one of the most popular Graph Neural Network (GNN) architecture, which employs the attention mechanism to learn edge weights and has demonstrated promising performance in various applications. However, since it only incorporates information from immediate neighborhood, it lacks the ability to capture long-range and global graph information, leading to unsatisfactory performance on some datasets, particularly on heterophilic graphs. To address this limitation, we propose the Directional Graph Attention Network (DGAT) in this paper. DGAT is able to combine the feature-based attention with the global directional information extracted from the graph topology. To this end, a new class of Laplacian matrices is proposed which can provably reduce the diffusion distance between nodes. Based on the new Laplacian, topology-guided neighbour pruning and edge adding mechanisms are proposed to remove the noisy and capture the helpful long-range neighborhood information. Besides, a global directional attention is designed to enable a topological-aware information propagation. The superiority of the proposed DGAT over the baseline GAT has also been verified through experiments on real-world benchmarks and synthetic data sets. It also outperforms the state-of-the-art (SOTA) models on 6 out of 7 real-world benchmark datasets.

We leverage the Multiplicative Weight Update (MWU) method to develop a decentralized algorithm that significantly improves the performance of dynamic time division duplexing (D-TDD) in small cell networks. The proposed algorithm adaptively adjusts the time portion allocated to uplink (UL) and downlink (DL) transmissions at every node during each scheduled time slot, aligning the packet transmissions toward the most appropriate link directions according to the feedback of signal-to-interference ratio information. Our simulation results reveal that compared to the (conventional) fixed configuration of UL/DL transmission probabilities in D-TDD, incorporating MWU into D-TDD brings about a two-fold improvement of mean packet throughput in the DL and a three-fold improvement of the same performance metric in the UL, resulting in the D-TDD even outperforming Static-TDD in the UL. It also shows that the proposed scheme maintains a consistent performance gain in the presence of an ascending traffic load, validating its effectiveness in boosting the network performance. This work also demonstrates an approach that accounts for algorithmic considerations at the forefront when solving stochastic problems.