Recent years have witnessed an explosive growth of AI models. The high cost of hosting AI services on GPUs and their demanding service requirements, make it timely and challenging to lower service costs and guarantee service quality. While spot instances have long been offered with a large discount, spot preemptions have discouraged users from using them to host model replicas when serving AI models. To address this, we introduce SkyServe, a system that efficiently serves AI models over a mixture of spot and on-demand replicas across regions and clouds. SkyServe intelligently spreads spot replicas across different failure domains (e.g., regions or clouds) to improve availability and reduce correlated preemptions, overprovisions cheap spot replicas than required as a safeguard against possible preemptions, and dynamically falls back to on-demand replicas when spot replicas become unavailable. We compare SkyServe with both research and production systems on real AI workloads: SkyServe reduces cost by up to 44% while achieving high resource availability compared to using on-demand replicas. Additionally, SkyServe improves P50, P90, and P99 latency by up to 2.6x, 3.1x, 2.7x compared to other research and production systems.

Spurred by the demand for interpretable models, research on eXplainable AI for language technologies has experienced significant growth, with feature attribution methods emerging as a cornerstone of this progress. While prior work in NLP explored such methods for classification tasks and textual applications, explainability intersecting generation and speech is lagging, with existing techniques failing to account for the autoregressive nature of state-of-the-art models and to provide fine-grained, phonetically meaningful explanations. We address this gap by introducing Spectrogram Perturbation for Explainable Speech-to-text Generation (SPES), a feature attribution technique applicable to sequence generation tasks with autoregressive models. SPES provides explanations for each predicted token based on both the input spectrogram and the previously generated tokens. Extensive evaluation on speech recognition and translation demonstrates that SPES generates explanations that are faithful and plausible to humans.

Traditional natural disaster response involves significant coordinated teamwork where speed and efficiency are key. Nonetheless, human limitations can delay critical actions and inadvertently increase human and economic losses. Agentic Large Vision Language Models (LVLMs) offer a new avenue to address this challenge, with the potential for substantial socio-economic impact, particularly by improving resilience and resource access in underdeveloped regions. We introduce DisasTeller, the first multi-LVLM-powered framework designed to automate tasks in post-disaster management, including on-site assessment, emergency alerts, resource allocation, and recovery planning. By coordinating four specialised LVLM agents with GPT-4 as the core model, DisasTeller autonomously implements disaster response activities, reducing human execution time and optimising resource distribution. Our evaluations through both LVLMs and humans demonstrate DisasTeller's effectiveness in streamlining disaster response. This framework not only supports expert teams but also simplifies access to disaster management processes for non-experts, bridging the gap between traditional response methods and LVLM-driven efficiency.

Collaborative Filtering (CF) methods dominate real-world recommender systems given their ability to learn high-quality, sparse ID-embedding tables that effectively capture user preferences. These tables scale linearly with the number of users and items, and are trained to ensure high similarity between embeddings of interacted user-item pairs, while maintaining low similarity for non-interacted pairs. Despite their high performance, encouraging dispersion for non-interacted pairs necessitates expensive regularization (e.g., negative sampling), hurting runtime and scalability. Existing research tends to address these challenges by simplifying the learning process, either by reducing model complexity or sampling data, trading performance for runtime. In this work, we move beyond model-level modifications and study the properties of the embedding tables under different learning strategies. Through theoretical analysis, we find that the singular values of the embedding tables are intrinsically linked to different CF loss functions. These findings are empirically validated on real-world datasets, demonstrating the practical benefits of higher stable rank, a continuous version of matrix rank which encodes the distribution of singular values. Based on these insights, we propose an efficient warm-start strategy that regularizes the stable rank of the user and item embeddings. We show that stable rank regularization during early training phases can promote higher-quality embeddings, resulting in training speed improvements of up to 66%. Additionally, stable rank regularization can act as a proxy for negative sampling, allowing for performance gains of up to 21% over loss functions with small negative sampling ratios. Overall, our analysis unifies current CF methods under a new perspective, their optimization of stable rank, motivating a flexible regularization method.

Exploring the predictive capabilities of language models in material science is an ongoing interest. This study investigates the application of language model embeddings to enhance material property prediction in materials science. By evaluating various contextual embedding methods and pre-trained models, including Bidirectional Encoder Representations from Transformers (BERT) and Generative Pre-trained Transformers (GPT), we demonstrate that domain-specific models, particularly MatBERT significantly outperform general-purpose models in extracting implicit knowledge from compound names and material properties. Our findings reveal that information-dense embeddings from the third layer of MatBERT, combined with a context-averaging approach, offer the most effective method for capturing material-property relationships from the scientific literature. We also identify a crucial "tokenizer effect," highlighting the importance of specialized text processing techniques that preserve complete compound names while maintaining consistent token counts. These insights underscore the value of domain-specific training and tokenization in materials science applications and offer a promising pathway for accelerating the discovery and development of new materials through AI-driven approaches.

Cross-domain recommendation (CDR) offers a promising solution to the data sparsity problem by enabling knowledge transfer across source and target domains. However, many recent CDR models overlook crucial issues such as privacy as well as the risk of negative transfer (which negatively impact model performance), especially in multi-domain settings. To address these challenges, we propose FedGCDR, a novel federated graph learning framework that securely and effectively leverages positive knowledge from multiple source domains. First, we design a positive knowledge transfer module that ensures privacy during inter-domain knowledge transmission. This module employs differential privacy-based knowledge extraction combined with a feature mapping mechanism, transforming source domain embeddings from federated graph attention networks into reliable domain knowledge. Second, we design a knowledge activation module to filter out potential harmful or conflicting knowledge from source domains, addressing the issues of negative transfer. This module enhances target domain training by expanding the graph of the target domain to generate reliable domain attentions and fine-tunes the target model for improved negative knowledge filtering and more accurate predictions. We conduct extensive experiments on 16 popular domains of the Amazon dataset, demonstrating that FedGCDR significantly outperforms state-of-the-art methods.

Recently, tool-augmented LLMs have gained increasing attention. Given an instruction, tool-augmented LLMs can interact with various external tools in multiple rounds and provide a final answer. However, previous LLMs were trained on overly detailed instructions, which included API names or parameters, while real users would not explicitly mention these API details. This leads to a gap between trained LLMs and real-world scenarios. In addition, most works ignore whether the interaction process follows the instruction. To address these issues, we constructed a training dataset called MGToolBench, which contains statement and category-level instructions to better reflect real-world scenarios. In addition, we propose ToolPlanner, a two-stage reinforcement learning framework that utilizes path planning and two feedback mechanisms to enhance the LLM's task completion and instruction-following capabilities. Experimental results show that ToolPlanner significantly improves the Match Rate, Pass Rate and Win Rate by 26.8%, 20.2%, and 5.6% compared to the SOTA model. Human evaluation verifies that the multi-granularity instructions can better align with users' usage habits. Our data and code will be released upon acceptance.

Evaluating text-to-image generative models remains a challenge, despite the remarkable progress being made in their overall performances. While existing metrics like CLIPScore work for coarse evaluations, they lack the sensitivity to distinguish finer differences as model performance rapidly improves. In this work, we focus on the text rendering aspect of these models, which provides a lens for evaluating a generative model's fine-grained instruction-following capabilities. To this end, we introduce a new evaluation framework called TypeScore to sensitively assess a model's ability to generate images with high-fidelity embedded text by following precise instructions. We argue that this text generation capability serves as a proxy for general instruction-following ability in image synthesis. TypeScore uses an additional image description model and leverages an ensemble dissimilarity measure between the original and extracted text to evaluate the fidelity of the rendered text. Our proposed metric demonstrates greater resolution than CLIPScore to differentiate popular image generation models across a range of instructions with diverse text styles. Our study also evaluates how well these vision-language models (VLMs) adhere to stylistic instructions, disentangling style evaluation from embedded-text fidelity. Through human evaluation studies, we quantitatively meta-evaluate the effectiveness of the metric. Comprehensive analysis is conducted to explore factors such as text length, captioning models, and current progress towards human parity on this task. The framework provides insights into remaining gaps in instruction-following for image generation with embedded text.

Recommender systems are widely used AI applications designed to help users efficiently discover relevant items. The effectiveness of such systems is tied to the satisfaction of both users and providers. However, user satisfaction is complex and cannot be easily framed mathematically using information retrieval and accuracy metrics. While many studies evaluate accuracy through offline tests, a growing number of researchers argue that online evaluation methods such as A/B testing are better suited for this purpose. We have employed a variety of algorithms on different types of datasets divergent in size and subject, producing recommendations in various platforms, including media streaming services, digital publishing websites, e-commerce systems, and news broadcasting networks. Notably, our target websites and datasets are in Persian (Farsi) language. This study provides a comparative analysis of a large-scale recommender system that has been operating for the past year across about 70 websites in Iran, processing roughly 300 requests per second collectively. The system employs user-based and item-based recommendations using content-based, collaborative filtering, trend-based methods, and hybrid approaches. Through both offline and online evaluations, we aim to identify where these algorithms perform most efficiently and determine the best method for our specific needs, considering the dataset and system scale. Our methods of evaluation include manual evaluation, offline tests including accuracy and ranking metrics like hit-rate@k and nDCG, and online tests consisting of click-through rate (CTR). Additionally we analyzed and proposed methods to address cold-start and popularity bias.

Next point-of-interest (POI) recommendation aims to predict a user's next destination based on sequential check-in history and a set of POI candidates. Graph neural networks (GNNs) have demonstrated a remarkable capability in this endeavor by exploiting the extensive global collaborative signals present among POIs. However, most of the existing graph-based approaches construct graph structures based on pre-defined heuristics, failing to consider inherent hierarchical structures of POI features such as geographical locations and visiting peaks, or suffering from noisy and incomplete structures in graphs. To address the aforementioned issues, this paper presents a novel Bi-level Graph Structure Learning (BiGSL) for next POI recommendation. BiGSL first learns a hierarchical graph structure to capture the fine-to-coarse connectivity between POIs and prototypes, and then uses a pairwise learning module to dynamically infer relationships between POI pairs and prototype pairs. Based on the learned bi-level graphs, our model then employs a multi-relational graph network that considers both POI- and prototype-level neighbors, resulting in improved POI representations. Our bi-level structure learning scheme is more robust to data noise and incompleteness, and improves the exploration ability for recommendation by alleviating sparsity issues. Experimental results on three real-world datasets demonstrate the superiority of our model over existing state-of-the-art methods, with a significant improvement in recommendation accuracy and exploration performance.