While dense retrieval models have been the standard for state-of-the-art information retrieval, their deployment is often constrained by high memory requirements and reliance on GPU accelerators for vector similarity search at scale. Learned sparse retrieval offers a compelling alternative by enabling efficient search via inverted indices, yet it has historically received less attention than dense approaches. In this paper, we introduce LACONIC, a family of learned sparse retrievers based on the Llama3 architecture (1B, 3B, and 8B). We propose a streamlined two-phase training curriculum consisting of (1) weakly supervised pre-finetuning to adapt causal LLMs for bidirectional contextualization and (2) high-signal finetuning using curated hard negatives. Our results demonstrate that LACONIC effectively bridges the performance gap with dense models: the 8B variant achieves a state-of-the-art 60.2 nDCG@10 on the MTEB Retrieval benchmark, ranking 15th on the leaderboard as of February 5th, 2026, while utilizing 74% less index memory than an equivalent dense model. By delivering high retrieval effectiveness on commodity CPU hardware with a fraction of the compute budget required by competing models, LACONIC provides a scalable and efficient solution for real-world search applications. We fully open source our code implementation and trained checkpoints to facilitate reproducibility.

Accurate chemical reaction prediction is essential for drug discovery and synthetic planning. However, this task becomes particularly challenging in low-data scenarios, where novel reaction types lack sufficient training examples. To address this challenge, we propose FewRxn, a novel model-agnostic few-shot reaction prediction framework that enables rapid adaptation to unseen reaction types using only a few training samples. FewRxn integrates several key innovations, including segmentation masks for enhanced reactant representation, fingerprint embeddings for richer molecular context, and task-aware meta-learning for effective knowledge transfer. Through extensive evaluations, FewRxn achieves state-of-the-art accuracy in few-shot settings, significantly outperforming traditional fine-tuning methods. Additionally, our work provides insights into the impact of molecular representations on reaction knowledge transfer, demonstrating that knowledge captured under molecular graph-based formulation consistently outperforms those learned in forms of SMILES generation in few-shot learning.

Transferring the reasoning capability from stronger large language models (LLMs) to smaller ones has been quite appealing, as smaller LLMs are more flexible to deploy with less expense. Among the existing solutions, knowledge distillation stands out due to its outstanding efficiency and generalization. However, existing methods suffer from several drawbacks, including limited knowledge diversity and the lack of rich contextual information. To solve the problems and facilitate the learning of compact language models, we propose TinyLLM, a new knowledge distillation paradigm to learn a small student LLM from multiple large teacher LLMs. In particular, we encourage the student LLM to not only generate the correct answers but also understand the rationales behind these answers. Given that different LLMs possess diverse reasoning skills, we guide the student model to assimilate knowledge from various teacher LLMs. We further introduce an in-context example generator and a teacher-forcing Chain-of-Thought strategy to ensure that the rationales are accurate and grounded in contextually appropriate scenarios. Extensive experiments on six datasets across two reasoning tasks demonstrate the superiority of our method. Results show that TinyLLM can outperform large teacher LLMs significantly, despite a considerably smaller model size. The source code is available at: https://github.com/YikunHan42/TinyLLM.

Current face recognition (FR) algorithms frequently encounter discrimination issues in terms of various attributes (e.g., gender, age) due to the biased demographic distribution of the training datasets towards specific groups. In this paper, we study an identity protected fair FR problem where the goal is to augment the datasets with external face images while ensuring the anonymity of the corresponding face identities. Our problem is motivated by the limitation of current fairness driven data augmentation approaches that directly utilize the external face images accessed by FR algorithm developers while ignoring the protection on the face identities of the image owners. To address the problem, we develop FaDE, a face segment driven identity anonymization framework that augments biased face image datasets by identifying specific face segments with diversified demographic characteristics from external face images but with least identity disclosure, and then reconstructing the segments to full face images with new identities. As a result, the augmented dataset is under a more balanced demographic distribution and improves the fairness performance of the optimized FR algorithms. We evaluate FaDE on two public face datasets, CelebA and LFW that suffer from various demographic imbalance. The results show that FaDE significantly enhances both fairness and accuracy performance of the optimized FR algorithms, while keeping effective anonymity for the identities of external face images.

Significant work has been conducted in the domain of food computing, yet these studies typically focus on single tasks such as t2t (instruction generation from food titles and ingredients), i2t (recipe generation from food images), or t2i (food image generation from recipes). None of these approaches integrate all modalities simultaneously. To address this gap, we introduce a novel food computing foundation model that achieves true multimodality, encompassing tasks such as t2t, t2i, i2t, it2t, and t2ti. By leveraging large language models (LLMs) and pre-trained image encoder and decoder models, our model can perform a diverse array of food computing-related tasks, including food understanding, food recognition, recipe generation, and food image generation. Compared to previous models, our foundation model demonstrates a significantly broader range of capabilities and exhibits superior performance, particularly in food image generation and recipe generation tasks. We open-sourced ChefFusion at https://github.com/Peiyu-Georgia-Li/ChefFusion-Multimodal-Foundation-Model-Integrating-Recipe-and-Food-Image-Generation.git.

The success of GNNs highly relies on the accurate labeling of data. Existing methods of ensuring accurate labels, such as weakly-supervised learning, mainly focus on the existing nodes in the graphs. However, in reality, new nodes always continuously emerge on dynamic graphs, with different categories and even label noises. To this end, we formulate a new problem, Graph Cross-Supervised Learning, or Graph Weak-Shot Learning, that describes the challenges of modeling new nodes with novel classes and potential label noises. To solve this problem, we propose Lipshitz-regularized Mixture-of-Experts similarity network (LIME), a novel framework to encode new nodes and handle label noises. Specifically, we first design a node similarity network to capture the knowledge from the original classes, aiming to obtain insights for the emerging novel classes. Then, to enhance the similarity network's generalization to new nodes that could have a distribution shift, we employ the Mixture-of-Experts technique to increase the generalization of knowledge learned by the similarity network. To further avoid losing generalization ability during training, we introduce the Lipschitz bound to stabilize model output and alleviate the distribution shift issue. Empirical experiments validate LIME's effectiveness: we observe a substantial enhancement of up to 11.34% in node classification accuracy compared to the backbone model when subjected to the challenges of label noise on novel classes across five benchmark datasets. The code can be accessed through https://github.com/xiangchi-yuan/Graph-Cross-Supervised-Learning.

Machine Unlearning (MU) algorithms have become increasingly critical due to the imperative adherence to data privacy regulations.The primary objective of MU is to erase the influence of specific data samples on a given model without the need to retrain it from scratch.Accordingly, existing methods focus on maximizing user privacy protection.However, there are different degrees of privacy regulations for each real-world web-based application.Exploring the full spectrum of trade-offs between privacy, model utility, and runtime efficiency is critical for practical unlearning scenarios.Furthermore, designing the MU algorithm with simple control of the aforementioned trade-off is desirable but challenging due to the inherent complex interaction.To address the challenges, we present Controllable Machine Unlearning (ConMU), a novel framework designed to facilitate the calibration of MU.The ConMU framework contains three integral modules: an important data selection module that reconciles the runtime efficiency and model generalization, a progressive Gaussian mechanism module that balances privacy and model generalization, and an unlearning proxy that controls the trade-offs between privacy and runtime efficiency.Comprehensive experiments on various benchmark datasets have demonstrated the robust adaptability of our control mechanism and its superiority over established unlearning methods.ConMU explores the full spectrum of the Privacy-Utility-Efficiency trade-off and allows practitioners to account for different real-world regulations.

Graph Neural Networks (GNNs) have demonstrated a great representation learning capability on graph data and have been utilized in various downstream applications. However, real-world data in web-based applications (e.g., recommendation and advertising) always contains bias, preventing GNNs from learning fair representations. Although many works were proposed to address the fairness issue, they suffer from the significant problem of insufficient learnable knowledge with limited attributes after debiasing. To address this problem, we develop Graph-Fairness Mixture of Experts (G-Fame), a novel plug-and-play method to assist any GNNs to learn distinguishable representations with unbiased attributes. Furthermore, based on G-Fame, we propose G-Fame++, which introduces three novel strategies to improve the representation fairness from node representations, model layer, and parameter redundancy perspectives. In particular, we first present the embedding diversified method to learn distinguishable node representations. Second, we design the layer diversified strategy to maximize the output difference of distinct model layers. Third, we introduce the expert diversified method to minimize expert parameter similarities to learn diverse and complementary representations. Extensive experiments demonstrate the superiority of G-Fame and G-Fame++ in both accuracy and fairness, compared to state-of-the-art methods across multiple graph datasets.

The spread of COVID-19 throughout the world has led to cataclysmic consequences on the global community, which poses an urgent need to accurately understand and predict the trajectories of the pandemic. Existing research has relied on graph-structured human mobility data for the task of pandemic forecasting. To perform pandemic forecasting of COVID-19 in the United States, we curate Large-MG, a large-scale mobility dataset that contains 66 dynamic mobility graphs, with each graph having over 3k nodes and an average of 540k edges. One drawback with existing Graph Neural Networks (GNNs) for pandemic forecasting is that they generally perform information propagation in a flat way and thus ignore the inherent community structure in a mobility graph. To bridge this gap, we propose a Hierarchical Spatio-Temporal Graph Neural Network (HiSTGNN) to perform pandemic forecasting, which learns both spatial and temporal information from a sequence of dynamic mobility graphs. HiSTGNN consists of two network architectures. One is a hierarchical graph neural network (HiGNN) that constructs a two-level neural architecture: county-level and region-level, and performs information propagation in a hierarchical way. The other network architecture is a Transformer-based model that captures the temporal dynamics among the sequence of learned node representations from HiGNN. Additionally, we introduce a joint learning objective to further optimize HiSTGNN. Extensive experiments have demonstrated HiSTGNN's superior predictive power of COVID-19 new case/death counts compared with state-of-the-art baselines.

Learning effective representations for recipes is essential in food studies for recommendation, classification, and other applications. Unlike what has been developed for learning textual or cross-modal embeddings for recipes, the structural relationship among recipes and food items are less explored. In this paper, we formalize the problem recipe representation learning with networks to involve both the textual feature and the structural relational feature into recipe representations. Specifically, we first present RecipeNet, a new and large-scale corpus of recipe data to facilitate network based food studies and recipe representation learning research. We then propose a novel heterogeneous recipe network embedding model, rn2vec, to learn recipe representations. The proposed model is able to capture textual, structural, and nutritional information through several neural network modules, including textual CNN, inner-ingredients transformer, and a graph neural network with hierarchical attention. We further design a combined objective function of node classification and link prediction to jointly optimize the model. The extensive experiments show that our model outperforms state-of-the-art baselines on two classic food study tasks. Dataset and codes are available at https://github.com/meettyj/rn2vec.