Large foundation models have demonstrated remarkable success in natural language processing and computer vision. Applying the large models to downstream tasks often requires fine-tuning, in order to boost the predictive accuracy. However, the fine-tuning process relies heavily on labeled data and extensive training. This dependency makes fine-tuning impractical for niche applications, such as rare object detection or specialized medical tasks. To overcome these limitations, we propose KALE: Knowledge Aggregation for Label-free model Enhancement, a label-free method for model enhancement, leveraging knowledge aggregation via model fusion and adaptive representation alignment. Our method is powered by a carefully designed joint self-cooperative optimization function that considers (i) multi-granularity optimization (task-specific and layer-specific), (ii) self and cooperative supervision integration, and (iii) mitigation of error accumulation caused by entropy minimization. Additionally, we introduce a class cardinality-aware sample filtering to ensure the stability of the fusion process. We also design a lightweight representation alignment technique to refine the fusion coefficient in a few shots for quality enhancement. We evaluate our method on multiple image classification datasets using ViT-B/32 and ViT-L/14 backbones. Experimental results demonstrate that our label-free method consistently outperforms state-of-the-art unsupervised approaches, including TURTLE and supervised full fine-tuning, in terms of average performance. Specifically, compared to TURTLE, our method achieves average improvements of 20.7% with ViT-B/32 and 19.5% with ViT-L/14. Furthermore, on the challenging SUN397 dataset, our method surpasses supervised full fine-tuning by 4% and 2.3% with ViT-B/32 and ViT-L/14, respectively.

Tuning inference hyperparameters, such as temperature and maximum output tokens, on downstream tasks can enhance inference performance.However, directly applying hyperparameter optimization to these hyperparameters is token-expensive.Multi-fidelity optimization improves HPO efficiency with lowfidelity evaluations, but its static scheduling strategies ignore token consumption, leading to high costs.To address these limitations, we propose a token-efficient multi-fidelity optimization method, which enhances inference performance and minimizes token usage.Our method is empowered by (i) a token-based fidelity definition with explicit token cost modeling on configurations; (ii) a novel Token-Aware Expected Improvement acquisition function that selects configurations based on performance gain per token; and (iii) a dynamic fidelity scheduling mechanism that adapts to real-time budget status.We evaluate our method on LLaMA-2 and LLaMA-3 series across MMLU, Humaneval, MedQA, and OpenBookQA.Our method improves over the HELM leaderboard by 7.1%, 24.3%, 21.9%, and 4.6%, respectively.Compared to existing multi-fidelity HPO baselines, our method reduces token consumption by over 80% while maintaining or surpassing performance, demonstrating the state-of-the-art token efficiency for inference-time optimization.

Due to the complexity and self-evolutionary property of graph data in reality, graph learning methods require both validity to represent unstructured data and scalability to adapt to evolving graphs. However, current works have representation learning limitations on optimizable graph feature space due to the bottleneck of the structure depth. Moreover, they encounter a complete retraining process when graphs evolve, especially in the case without the assistance of new labels. To address the above issues, we propose a scalable graph broad network (SGB-Net), which contains three proposed modules: the graph feature broad transformation layer (GFBT layer) for enhancing graph embedding and two update algorithms (SGB-Net-U, SGB-Net-S) for endowing scalability. The GFBT layer aims to explicitly expand the graph feature space and broadly build the model. It constructs two expandable feature spaces in various graph scales to embed graphs discriminatively. SGB-Net-U is an exploratory method designed to tackle the label-free graph incremental learning (GIL) problem by leveraging unsupervised incremental knowledge to expand graph representation. SGB-Net-S endows scalability in classical incremental learning scenarios involving labels. Benefiting from its broad construction framework, SGB-Net not only enhances graph embeddings but also seamlessly adapts and improves performance in response to graph expansion without requiring retraining. In the experiments conducted on 15 benchmark datasets, SGB-Net outperforms state-of-the-art GNNs in terms of both effectiveness and scalability.

Sleep stage classification assists in evaluating sleep quality and diagnosing sleep disorders. Existing methods based on spatial-temporal graph convolutional networks (STGCNs) have shown excellent performance in recognizing sleep stages. However, current works learn transition dependencies in sleep sequence in a coarse manner and restrict the modeling of brain activity to a significantly brief period. To address the above issues, we propose a fine-grained sleep transition graph (FSTG) method to learn EEG sequence dependencies for sleep stage classification in this paper. Specifically, the novel sleep transition graph (STG) constructor (STG-C) is proposed to model sleep sequence transition dependencies in a more fine-grained manner, extracting effective auxiliary information from sleep contextual sequences for sleep staging. In the STG-C, we initially explore brain connectivity over extended sleep temporal scales, utilizing the graph fusion mechanism (GFM) to learn the multiple temporal correlations of brain channels. Furthermore, to better capture dependencies of sleep sequence, a skip-gated STGCN (SG-STGCN) is designed as the strong STG sequence learner. The proposed FSTG model achieves state-of-the-art performance in public sleep staging datasets, ISRUC-S3 and ISRUC-S1. Additionally, visualization experiments demonstrate the effectiveness and interpretability of FSTG in sleep staging classification.

Supervised learning is a powerful tool that has shown promise when applied towards bioinformatics data sets. Deep forest, a supervised ensemble model based on decision trees, has been proven to have excellent classification performance and strong generalization ability across different fields. However, when dealing with high-dimensional and small-sample gene expression data, commonly used supervised learning methods including deep forest may not be effective. In this paper, we propose DPCAForest, a deep-forest-based model, which integrates deep forest and dynamic principle component analysis. DPCAForest adaptively generates the minority samples based on sample distribution, then conducts principle component analysis dynamically synchronized with growth of deep forest to reveal the important features with the highest variance. Dynamic PCA enables the model to perform feature extraction in a data-driven way based on cross-validation, and the model can obtain fusion information across layers. In the experimental studies, DPCAForest is verified on Adenocarcinoma, Brain, Colon, Small Round Blue Cell Tumors (SRBCTs) and NCI-60 cancer data sets and demonstrated desirable or better performance than state-of-the-art methods in terms of accuracy and F-Measure.

The user and services provided by resources in grid environments are dynamic. Therefore, the malicious nodes may exist in grid system. It will significantly affect the quality of service (QOS) requirements of users, even lead to the economic loss of user in the grid economic environment. So a certain resource management scheme has to be implemented in grid economy environments to ensure the environments working properly and achieve better scalability. It is recognized an effective measure which uses Reputation to control the access of malicious nodes. In this paper, we design a distributed reputation control architecture which is based on virtual organizational domains (VOD) and propose a reputation calculation algorithm which spans these domains. The inner structure and this framework are also detailed discussed here. At last, through the simulating experiments, it can be proved that this architecture could block the assessment of malicious nodes effectively and improve the efficient and stability of nodes in grid economy environments.