Federated Learning (FL) is an emerging distributed learning paradigm under privacy constraint. Data heterogeneity one of the main challenges in FL, which results slow convergence and degraded performance. Most existing approaches only tackle challenge by restricting local model update client, ignoring performance drop caused direct global aggregation. Instead, we propose a data-free knowledge distillation method to fine-tune server (FedFTG), relieves issue Concretely, FedFTG explores input...

Time series forecasting is essential for a wide range of real-world applications. Recent studies have shown the superiority Transformer in dealing with such problems, especially long sequence time input (LSTI) and (LSTF) problems. To improve efficiency enhance locality Transformer, these combine CNN varying degrees. However, their combinations are loosely-coupled do not make full use CNN. address this issue, we propose concept tightly-coupled convolutional (TCCT) three TCCT architectures...

This paper shows that time series forecasting Transformer (TSFT) suffers from severe over-fitting problem caused by improper initialization method of unknown decoder inputs, especially when handling non-stationary series. Based on this observation, we propose GBT, a novel two-stage framework with Good Beginning. It decouples the prediction process TSFT into two stages, including Auto-Regression stage and Self-Regression to tackle different statistical properties between input sequences....

Label differential privacy (DP) is designed for learning problems involving private labels and public features. While various methods have been proposed under label DP, the theoretical limits remain largely unexplored. In this paper, we investigate fundamental of with DP in both local central models classification regression tasks, characterized by minimax convergence rates. We establish lower bounds converting each task into a multiple hypothesis testing problem bounding test error....

Predicting the trajectory of clinical decline in aging individuals is a pressing challenge, especially for people with mild cognitive impairment, Alzheimer′s disease, Parkinson′s or vascular dementia. Accurate predictions can guide treatment decisions, identify risk factors, and optimize trials. In this study, we compared two deep learning approaches forecasting changes, over 2-year interval, Clinical Dementia Rating scale ′sum boxes′ score (sobCDR). This key metric dementia research, scores...

Alzheimer's disease (AD) is a complex disorder that affects multiple biological systems including cognition, behavior and physical health. Unfortunately, the pathogenic mechanisms behind AD are not yet clear treatment options still limited. Despite increasing number of studies examining pairwise relationships between genetic factors, activity (PA), AD, few have successfully integrated all three domains data, which may help reveal impact these genomic phenomic factors on AD. We use...

Abstract Motivation The integration of Machine Learning (ML) and Artificial Intelligence (AI) into healthcare has immense potential due to the rapidly growing volume clinical data. However, existing AI models, particularly Large Language Models (LLMs) like GPT-4, face significant challenges in terms explainability reliability, high-stakes domains healthcare. Results This paper proposes a novel LLM-aided feature engineering approach that enhances interpretability by extracting clinically...

A new era of e-learning is on the horizon, hundreds learning contents are created and more people begin to acquire acknowledge through e-learning. The traditional teaching method already showing its limitations that students from different backgrounds still given same at time, they may only interest in part a whole content. In this paper, we propose novel way organise into small 'atomic' units called objects so could be used reused effectively. together with their ontology systemised...

With the development of Internet Things (IoT) systems, precise long-term forecasting method is requisite for decision makers to evaluate current statuses and formulate future policies. Currently, Transformer MLP are two paradigms deep time-series former one more prevailing in virtue its exquisite attention mechanism encoder-decoder architecture. However, data scientists seem be willing dive into research encoder, leaving decoder unconcerned. Some researchers even adopt linear projections...

This paper presents FDNet: a Focal Decomposed Network for efficient, robust and practical time series forecasting. We break away from conventional deep forecasting formulas which obtain prediction results universal feature maps of input sequences. In contrary, FDNet neglects correlations elements only extracts fine-grained local features sequence. show that: (1) Deep with sequence is feasible upon theoretical basis. (2) By abandoning global coarse-grained maps, overcomes distribution shift...

Parkinson's disease (PD) is the second most prevalent neurodegenerative in United States. The structural or functional connectivity between regions of interest (ROIs) brain and their changes captured connectomes could be potential biomarkers for PD. To effectively model complex non-linear characteristic connectomic patterns related to PD exploit long-range feature interactions ROIs, we propose a connectome transformer patient classification biomarker identification. proposed learns key by...

Morphometricity examines the global statistical association between brain morphology and an observable trait, is defined as proportion of trait variation attributable to morphology. In this work, we propose accurate morphometricity estimator based on generalized random effects (GRE) model, perform analyses five cognitive traits in Alzheimer's study. Our empirical study shows that proposed GRE model outperforms widely used LME both simulation real data. addition, extend estimation from whole...

Abstract Background Studies have shown physical activity (PA) patterns are heritable traits and correlated with several known genetic risk factors including APOE, the best‐known gene associated Alzheimer’s Disease (AD). SPARE‐AD was a previously developed machine learning index to be sensitive AD‐like brain atrophy. However, relationship between variants, AD‐related neuroimaging features yet been extensively studied due lack of appropriate data statistical methods for handling complex...

Abstract Background Alzheimer’s disease (AD) is a progressive neurological disorder with an unclear cause, and its amyloid tau hypotheses need deeper investigation to understand their link genetic variants AD outcomes. To bridge this gap, we conduct mediation analysis using the genotyping, imaging, cognitive data from Disease Neuroimaging Initiative (ADNI) delineate specific pathways such as single nucleotide polymorphisms (SNPs), regional amyloid/tau protein aggregation in brain, Method...

Abstract Background The amyloid‐tau‐neurodegeneration (ATN) framework provides a valuable model for comprehending the pathophysiology and progression of Alzheimer’s disease (AD). However relationship between genetic interaction with these three characteristics are complex not fully understood. Here, we use neuroimaging‐derived quantitative traits to evaluate risk amyloid accumulation, tau pathology, neurodegeneration. Method Disease Sequencing Project (ADSP) collected harmonized whole genome...

Privacy protection of users' entire contribution samples is important in distributed systems. The most effective approach the two-stage scheme, which finds a small interval first and then gets refined estimate by clipping into interval. However, operation induces bias, serious if sample distribution heavy-tailed. Besides, users with large local sizes can make sensitivity much larger, thus method not suitable for imbalanced users. Motivated these challenges, we propose Huber loss minimization...

User-level privacy is important in distributed systems. Previous research primarily focuses on the central model, while local models have received much less attention. Under user-level DP strictly stronger than item-level one. However, under relationship between and LDP becomes more complex, thus analysis crucially different. In this paper, we first analyze mean estimation problem then apply it to stochastic optimization, classification, regression. particular, propose adaptive strategies...

Recent advancements in offline reinforcement learning (RL) have underscored the capabilities of Conditional Sequence Modeling (CSM), a paradigm that learns action distribution based on history trajectory and target returns for each state. However, these methods often struggle with stitching together optimal trajectories from sub-optimal ones due to inconsistency between sampled within individual across multiple trajectories. Fortunately, Dynamic Programming (DP) offer solution by leveraging...

Personalized Federated Learning (PFL) is proposed to find the greatest personalized models for each client. To avoid central failure and communication bottleneck in server-based FL, we concentrate on Decentralized (DPFL) that performs distributed model training a Peer-to-Peer (P2P) manner. Most works DPFL are based undirected symmetric topologies, however, data, computation resources heterogeneity result large variances models, which lead aggregation suboptimal performance unguaranteed...

Foundation models have demonstrated significant emergent abilities, holding great promise for enhancing embodied agents' reasoning and planning capacities. However, the absence of a comprehensive benchmark evaluating agents with multimodal observations in complex environments remains notable gap. In this paper, we present MuEP, Multimodal Embodied Planning. MuEP facilitates evaluation multi-turn interactions scenes, incorporating fine-grained metrics that provide insights into performance...

Deep-learning-based methods have achieved promising performance in visual tracking tasks. However, the backbones of existing trackers normally emanate from object detection realm, making them inefficient and insufficient terms spatial template matching. Moreover, such apply temporal information without considering its authenticity during online inference step, rendering prone to error accumulation. To address these two issues, this work proposes OTETrack, a novel tracker with overlapped...