Multilingual large language-vision models (LVLMs), which understand and generate both text and images across multiple languages, have achieved remarkable performance on English-centric multimodal generation tasks. However, their performance on non-English tasks has been underwhelming. One major challenge with multilingual LVLMs is the modality gap between visual inputs and multilingual textual inputs/outputs due to the lack of high-quality multilingual training data. In this paper, we propose LRM-LLaVA, a multilingual large language-vision model designed for low-resource languages to overcome the modality gap. It is composed of four components: a visual encoder, a multilingual large language model, a vision-text representation projector, and a cross-modal regularizer. Both the projector and regularizer aim at reducing the modality gap and improving multilingual performance. To train LRM-LLaVA, we employ a two-stage training strategy including pre-training and instruction fine-tuning. Meanwhile, we construct a multilingual visual question answering dataset based on English open-source datasets and adopt multiple task instructions. To evaluate the performance of LVLMs across various languages, we construct four multilingual benchmarks for 10 languages, based on English open-source benchmarks. Experimental results show that LRM-LLaVA achieves competitive performance compared to other multilingual LVLMs of similar parameters.

Retrieval-augmented generation (RAG) greatly enhances large language models (LLMs) performance in knowledge-intensive tasks. However, naive RAG methods struggle with multi-hop question answering due to their limited capacity to capture complex dependencies across documents. Recent studies employ graph-based RAG to capture document connections. However, these approaches often result in a loss of semantic coherence and introduce irrelevant noise during node matching and subgraph construction. To address these limitations, we propose NeuroPath, an LLM-driven semantic path tracking RAG framework inspired by the path navigational planning of place cells in neurobiology. It consists of two steps: Dynamic Path Tracking and Post-retrieval Completion. Dynamic Path Tracking performs goal-directed semantic path tracking and pruning over the constructed knowledge graph (KG), improving noise reduction and semantic coherence. Post-retrieval Completion further reinforces these benefits by conducting second-stage retrieval using intermediate reasoning and the original query to refine the query goal and complete missing information in the reasoning path. NeuroPath surpasses current state-of-the-art baselines on three multi-hop QA datasets, achieving average improvements of 16.3\% on recall@2 and 13.5\% on recall@5 over advanced graph-based RAG methods. Moreover, compared to existing iter-based RAG methods, NeuroPath achieves higher accuracy and reduces token consumption by 22.8\%. Finally, we demonstrate the robustness of NeuroPath across four smaller LLMs (Llama3.1, GLM4, Mistral0.3, and Gemma3), and further validate its scalability across tasks of varying complexity. Code is available at [https://github.com/KennyCaty/NeuroPath](https://github.com/KennyCaty/NeuroPath).

With the development of AI model ecology pioneered by the large language models, data plays a more important and diversified role than the traditional one. In recent years, data problems have been increasingly noticed and voiced by developers, and it is crucial to understand the existing challenges that they are facing in reality, as well as the potential features of these problems. Unfortunately, we currently have relatively little knowledge in this area. To fill this gap, we conduct an empirical study based on 2,154 posts and 785 issues by collecting model data problems from three online communities (i.e., GitHub, Stack Overflow, and Hugging Face). We present the first taxonomy of data problems around model production encountered by developers, covering 11 topics in 4 categories. Then, we analyze the popularity and the difficulty of the posts and issues in the derived topics. We distill many findings from the study and present the practical implications from the perspectives of different audiences.

With more attention paid to communication security and user privacy, there has been a dramatic rise in private protocols, which brings great challenges to traditional protocol identification. The increasingly complex design of private pro-tocols has led to the loss of effectiveness of features used for traditional identification. Besides, some features are redundant and also lose sight of the particularity of private protocols, which could not be guaranteed to be the optimal identification solution. In this paper, we propose a method to obtain the Sequence Sta-tistical Fingerprint(Seq-SF) which is proprietary for each private protocol and can be used for identifying private protocols from promiscuous traffic. We extend statistical features associated with protocol transmission to make the sequence feature be integrated into it. Then, we adopt a feature selection algorithm based on mutual information ranking, which can evaluate the pertinence between features and protocols to guarantee the effectiveness and optimality of Seq-SF. The results of experiments on the real-world dataset covering 13 private protocols indicate that Seq-SF achieves an ideal performance(99.41% accuracy, 97.56% precision, 95.18% recall, and 0.20% FPR on average) and has good robustness.

High-fidelity prototyping tools are used by software designers and developers to iron out interface details without full implementation. However, the lack of visual accessibility in these tools creates a barrier for designers who may use screen readers, such as those who are vision impaired. We assessed conformance of four prototyping tools (Sketch, Adobe XD, Balsamiq, UXPin) with accessibility guidelines, using two screen readers (Narrator and VoiceOver), focusing our analysis on GUI element accessibility and critical workflows used to create prototypes. We found few tools were fully accessible, with 45.9% of GUI elements meeting accessibility criteria (34.2% partially supported accessibility, 19.9% not supporting accessibility). Accessibility issues stymied efforts to create prototypes using screen readers. Though no screen reader-tool pairs were completely accessible, the most accessible pairs were VoiceOver-Sketch, VoiceOver-Balsamiq, and Narrator-Balsamiq. We recommend prioritizing improved accessibility for input and control instruction, alternative text, focus order, canvas element properties, and keyboard operations.

Building structure and other factors lead to the performance deterioration of global postioning system (GPS) positioning systems indoors. An adaptive model for Bluetooth-based indoor positioning is proposed in this paper, targeting at the complex indoor environment, to improve the performance of Bluetooth-oriented indoor positioning systems. More accurate Received Signal Strength Indicator (RSSI) calibration which is optimized via Gaussian filtering, together with the environment-dependent attenuation coefficient optimization, results in a more precise hybrid model in the complicated indoor environment. Experiment results show that the difference between the estimated results and the measured samples is less than 0.25[Formula: see text]m as the target node and reference node is less than 1.5[Formula: see text]m far from each other. As the distance increases to more than 1.5[Formula: see text]m, the relative difference between the estimated values and the measured ones decreases to 7.8% at most, satisfying the requirements for indoor positioning applications.

This paper presents a torque distribution algorithm aiming to improve both the energy economy and vehicle performance for electric vehicles with four in-wheel motors. To incorporate energy management issue into the torque distribution problem, a mathematical energy efficiency model for the driving motor is built based on experimental data. The design of the multi-objective optimization problem considering the following three factors: energy consumption, road handling and ride comfortability. To solve this complex objective function, an Adaptive Particle Swarm Optimization (APSO) algorithm is adopted to figure out this multi-objective optimization problem. Co-simulation based on CarSim and Simulink indicates that the proposed torque distribution algorithm can improve energy economy, road handling and ride comfortability of the electric vehicle with four in-wheel motors compared with the even-distribution algorithm.