Scene perception and trajectory forecasting are two fundamental challenges that crucial to a safe reliable autonomous driving (AD) system. However, most proposed methods aim at addressing one of the mentioned above with single model. To tackle this dilemma, paper proposes spatio-temporal semantics interaction graph aggregation for multi-agent (ST-SIGMA), an efficient end-to-end method jointly accurately perceive AD environment forecast trajectories surrounding traffic agents within unified...

Causal feature selection has achieved much attention in recent years, which discovers a Markov boundary (MB) of the class attribute. The MB attribute implies local causal relations between and features, thus leading to more interpretable robust prediction models than features selected by traditional algorithms. Many methods have been proposed, almost all them employ conditional independence (CI) tests identify MBs. However, many datasets from real-world applications may suffer incorrect CI...

Semi-supervised learning (SSL) concerns the problem of how to improve classifiers’ performance through making use prior knowledge from unlabeled data. Many SSL methods have been developed integrate data into classifiers based on either manifold or cluster assumption in recent years. In particular, graph-based approaches, following assumption, achieved a promising many real-world applications. However, most them work well small-scale sets only and lack probabilistic outputs. this paper,...

To cluster data that are not linearly separable in the original feature space, k -means clustering was extended to kernel version. However, performance of largely depends on choice function. mitigate this problem, multiple learning has been introduced into obtain an optimal combination for clustering. Despite success various scenarios, few existing work update coefficients based diversity kernels, which leads result selected kernels contain high redundancy and would degrade efficiency. We...

As data sources become ever more numerous with increased feature dimensionality, selection for multiview has an important technique in machine learning. Semi-supervised (SMFS) focuses on the problem of how to obtain a discriminative subset from heterogeneous spaces case abundant unlabeled little labeled data. Most existing methods suffer unreliable similarity graph structure across different views since they separate construction and use fixed graphs that are susceptible noisy features....

Recent years have witnessed the proliferation of techniques for streaming data mining to meet demands many real-time systems, where high-dimensional are generated at high speed, increasing burden on both hardware and software. Some feature selection algorithms proposed tackle this issue. However, these do not consider distribution shift due nonstationary scenarios, leading performance degradation when underlying changes in stream. To solve problem, article investigates through incremental...

Multi-label feature selection has received considerable attentions during the past decade. However, existing algorithms do not attempt to uncover underlying causal mechanism, and individually solve different types of variable relationships, ignoring mutual effects between them. Furthermore, these lack interpretability, which can only select features for all labels, but cannot explain correlation a selected certain label. To address problems, in this paper, we theoretically study...

As data with diverse representations become high-dimensional, multi-view unsupervised feature selection has been an important learning paradigm. Generally, existing methods encounter the following challenges: (i) traditional solutions either concatenate different views or introduce extra parameters to weight them, affecting performance and applicability; (ii) emphasis is typically placed on graph construction, yet disregarding clustering information of data; (iii) exploring similarity...

The explosive growth of text data requires effective methods to represent and classify these texts. Many learning have been proposed, like statistics-based methods, semantic similarity deep methods. focus on comparing the substructure text, which ignores between different words. Semantic learn a representation by training word embedding representing as average vector all However, cannot capture topic diversity words texts clearly. Recently, such CNNs RNNs studied. vanishing gradient problem...

With the increasing data dimensionality, feature selection has become a fundamental task to deal with high-dimensional data. Semi-supervised focuses on problem of how learn relevant subset in case abundant unlabeled few labeled In recent years, many semi-supervised algorithms have been proposed. However, these are implemented by separating processes and classifier training, such that they cannot simultaneously select features selected features. Moreover, ignore difference reliability inside...

Markov boundary (MB) has been widely studied in single-target scenarios. Relatively few works focus on the MB discovery for variable set due to complex relationships, where an might contain predictive information about several targets. This paper investigates multi-target discovery, aiming distinguish common variables (shared by multiple targets) and target-specific (associated with single targets). Considering multiplicity of MB, relation between equivalent is studied. We find that are...

The probabilistic classification vector machine (PCVM) synthesizes the advantages of both support and relevant machine, delivering a sparse Bayesian solution to problems. However, PCVM is currently only applicable binary cases. Extending multiclass cases via heuristic voting strategies such as one-vs-rest or one-vs-one often results in dilemma where classifiers make contradictory predictions, those might lose benefits outputs. To overcome this problem, we extend propose (mPCVM). Two learning...

As a foundational clustering paradigm, Density Peak Clustering (DPC) partitions samples into clusters based on their density peaks, garnering widespread attention. However, traditional DPC methods usually focus high-density regions, neglecting representative peaks in relatively low-density areas, particularly datasets with varying densities and multiple peaks. Moreover, existing variants struggle to identify correctly high-dimensional spaces due the indistinct distance differences among...

As partial samples are often absent in certain views, incomplete multi-view clustering has become a challenging task. To tackle data with missing current methods either utilize the similarity relations to recover or primarily consider available information of existing samples, typically facing some inherent limitations. Firstly, traditional solutions cannot fully explore potential contained due their omission strategy, leading sub-optimal graphs. Moreover, most mainly focus on recovery from...

Algorithm selection, a critical process of automated machine learning, aims to identify the most suitable algorithm for solving specific problem prior execution. Mainstream selection techniques heavily rely on features, while role features remains largely unexplored. Due intrinsic complexity algorithms, effective methods universally extracting information are lacking. This paper takes significant step towards bridging this gap by introducing Large Language Models (LLMs) into first time. By...

Negative correlation learning (NCL) is an ensemble algorithm that introduces a penalty term to the cost function of each individual member. Each member minimizes its mean square error and with rest ensemble. This paper analyzes NCL reveals adopting negative for unlabeled data beneficial improving model performance in semisupervised (SSL) setting. We then propose novel SSL algorithm, Semisupervised (SemiNCL) algorithm. The considers terms both labeled problems. In order reduce computational...

PRML, a novel candlestick pattern recognition model using machine learning methods, is proposed to improve stock trading decisions. Four popular methods and 11 different features types are applied all possible combinations of daily patterns start the schedule. Different time windows from one ten days used detect prediction effect at periods. An investment strategy constructed according identified suitable window. We deploy PRML for forecast Chinese market stocks Jan 1, 2000 until Oct 30,...

Sparse Bayesian learning is a state-of-the-art supervised algorithm that can choose subset of relevant samples from the input data and make reliable probabilistic predictions. However, in presence high-dimensional with irrelevant features, traditional sparse classifiers suffer performance degradation low efficiency due to incapability eliminating features. To tackle this problem, we propose novel embedded feature selection adopts truncated Gaussian distributions as both sample priors. The...