Cross-task Consistency Research Articles

Legal Judgment Prediction via graph boosting with constraints

Legal Judgment Prediction (LJP) is a multi-task multi-label problem in the civil law system, involving the prediction of law articles, charges, and terms of penalty based on fact descriptions. However, most existing research approaches LJP as a single-label scenario, neglecting the correlations between multiple labels and failing to consider cross-task consistency constraints in a multi-label scenario. Moreover, although previous multi-task studies have proposed expert models and coarse-grained topology construction for inter-task relationships, the former neglects rich information exchange among different tasks, and the latter, if one task’s prediction is inaccurate, will affect subsequent tasks. This paper has designed legal label graphs and proposed a novel graph boosting with constraints framework, GJudge, for legal judgment prediction to address these limitations. The framework comprises a multi-perspective interactive encoder and a multi-graph attention consistency expert module. The encoder utilizes bidirectional LSTM, gated attention units, cross attention, and graph attention networks to integrate fact descriptions and label similarity relationships information from legal label graphs for multi-perspective interactive encoding. The expert module utilizes the multiple expert networks and the multi-graph attention network to differentiate between confusing labels and ensure consistent constraints across tasks, this is achieved through the fusion of label consistency constraints and confusion relationships information in the legal label graphs. Experimental results on two real-world datasets across different tasks show an improvement in F1 scores ranging from at least 0.93% to a maximum of 2.97%, illustrating the effectiveness of GJudge compared to the state-of-the-art model.

TriangleNet: Edge Prior Augmented Network for Semantic Segmentation through Cross-Task Consistency

This paper addresses the task of semantic segmentation in computer vision, aiming to achieve precise pixel-wise classification. We investigate the joint training of models for semantic edge detection and semantic segmentation, which has shown a promise. However, implicit cross-task consistency learning in multitask networks is limited. To address this, we propose a novel “decoupled cross-task consistency loss” that explicitly enhances cross-task consistency. Our semantic segmentation network, TriangleNet, achieves a substantial 2.88% improvement over the Baseline in mean Intersection over Union (mIoU) on the Cityscapes test set. Notably, TriangleNet operates at 77.4% mIoU/46.2 FPS on Cityscapes, showcasing real-time inference capabilities at full resolution. With multiscale inference, performance is further enhanced to 77.8%. Furthermore, TriangleNet consistently outperforms the Baseline on the FloodNet dataset, demonstrating its robust generalization capabilities. The proposed method underscores the significance of multitask learning and explicit cross-task consistency enhancement for advancing semantic segmentation and highlights the potential of multitasking in real-time semantic segmentation.

Efficient simulation of CO2 migration dynamics in deep saline aquifers using a multi-task deep learning technique with consistency

CO2 sequestration and storage in deep saline aquifers is a promising technology for mitigating the excessive concentration of the greenhouse gas in the atmosphere. However, accurately predicting the migration of CO2 plumes requires complex multi-physics-based numerical simulation approaches, which are prohibitively expensive due to highly nonlinear coupled governing equations and uncertainties in heterogeneous spatial parameter distributions. To address this challenge, we developed an end-to-end deep learning workflow employing encoder–decoder architectures with residual network (ResNet) to efficiently predicts the spatial–temporal evolution of the solution CO2-brine ratio (Rs) and gas saturation (Sg) – the two essential tasks for quantifying the amount of trapped CO2 – given heterogeneous permeability fields as input. Specifically, we introduce a general multi-task learning with consistency (MTLC) framework to simultaneously predict Rs and Sg. The MTLC model leverages related tasks with less computational expensive labeled datasets to improve generalization ability. In our study, predictions for multiple tasks from the same permeability realization are not independent but expected to be consistent, as the proposed framework utilizes data-driven cross-task consistency constraints to augment learning of related tasks. Our deep learning model is trained based on physical trapping mechanisms, which play a dominant role in the CO2 migration process. The results demonstrate that the MTLC model with joint learning yields more accurate predictions and improved generalization for predicting CO2 migration in several test cases. Furthermore, our workflow is 105 times faster than a high-fidelity physics-based numerical simulator, making it a viable alternative for field-scale applications.

Uncertainty-guided mutual consistency learning for semi-supervised medical image segmentation

Medical image segmentation is a fundamental and critical step in many clinical approaches. Semi-supervised learning has been widely applied to medical image segmentation tasks since it alleviates the heavy burden of acquiring expert-examined annotations and takes the advantage of unlabeled data which is much easier to acquire. Although consistency learning has been proven to be an effective approach by enforcing an invariance of predictions under different distributions, existing approaches cannot make full use of region-level shape constraint and boundary-level distance information from unlabeled data. In this paper, we propose a novel uncertainty-guided mutual consistency learning framework to effectively exploit unlabeled data by integrating intra-task consistency learning from up-to-date predictions for self-ensembling and cross-task consistency learning from task-level regularization to exploit geometric shape information. The framework is guided by the estimated segmentation uncertainty of models to select out relatively certain predictions for consistency learning, so as to effectively exploit more reliable information from unlabeled data. Experiments on two publicly available benchmark datasets showed that: (1) Our proposed method can achieve significant performance improvement by leveraging unlabeled data, with up to 4.13% and 9.82% in Dice coefficient compared to supervised baseline on left atrium segmentation and brain tumor segmentation, respectively. (2) Compared with other semi-supervised segmentation methods, our proposed method achieve better segmentation performance under the same backbone network and task settings on both datasets, demonstrating the effectiveness and robustness of our method and potential transferability for other medical image segmentation tasks.

Deep learning of monocular depth, optical flow and ego-motion with geometric guidance for UAV navigation in dynamic environments

Computer vision-based depth estimation and visual odometry provide perceptual information useful for robot navigation tasks like obstacle avoidance. However, despite the proliferation of state-of-the-art convolutional neural network (CNN) models for monocular depth, ego-motion and optical flow estimation, a relatively low volume of work has been reported on their practical applications in unmanned aerial vehicle (UAV) navigation. This is due to well-known challenges — embedded hardware constraints, viewpoint variations, scarcity of aerial image datasets, and intricacies of dynamic environments. We address these limitations to facilitate real-world deployment of CNN in UAV navigation. First, we devise efficient confidence weighted adaptive network (Cowan) training framework that iteratively leverages intermediate prediction confidences to enforce cross-task consistency over corresponding image regions. This achieves competitive accuracy with a lightweight CNN capable of real-time execution on resource-constrained embedded systems. Second, we devise a test-time refinement method that adapts the network to dynamic environments while simultaneously improving accuracy. To accomplish this, we first update ego-motion using pose information from on-board inertial measurement unit (IMU). Then, we decompose the UAV’s motion into constituent vectors, and for each axis, we formulate geometric relationships between depth and translation. Based on this information, we triangulate corresponding points acquired through optical flow. Finally, we enforce geometric consistency between the initially updated pose and triangulated depth. Cowan with geometric guided refinement (Cowan-GGR) achieves significant accuracy and robustness. Field tests show the proposed model is capable of accurate depth and object-level motion perception in real-world dynamic environments, thus proving its efficacy in facilitating UAV navigation.

Cross-Task Consistency of Electroencephalography-Based Mental Workload Indicators: Comparisons Between Power Spectral Density and Task-Irrelevant Auditory Event-Related Potentials.

Mental workload (MWL) estimators based on ongoing electroencephalography (EEG) and event-related potentials (ERPs) have shown great potentials to build adaptive aiding systems for human–machine systems by estimating MWL in real time. However, extracting EEG features which are consistent in indicating MWL across different tasks is still one of the critical challenges. This study attempts to compare the cross-task consistency in indexing MWL variations between two commonly used EEG-based MWL indicators, power spectral density (PSD) of ongoing EEG and task-irrelevant auditory ERPs (tir-aERPs). The verbal N-back and the multi-attribute task battery (MATB), both with two difficulty levels, were employed in the experiment, along with task-irrelevant auditory probes. EEG was recorded from 17 subjects when they were performing the tasks. The tir-aERPs elicited by the auditory probes and the relative PSDs of ongoing EEG between two consecutive auditory probes were extracted and statistically analyzed to reveal the effects of MWL and task type. Discriminant analysis and support vector machine were employed to examine the generalization of tir-aERP and PSD features in indexing MWL variations across different tasks. The results showed that the amplitudes of tir-aERP components, N1, early P3a, late P3a, and the reorienting negativity, significantly decreased with the increasing MWL in both N-back and MATB. Task type had no obvious influence on the amplitudes and topological layout of the MWL-sensitive tir-aERP features. The relative PSDs in θ, α, and low β bands were also sensitive to MWL variations. However, the MWL-sensitive PSD features and their topological patterns were significantly affected by task type. The cross-task classification results based on tir-aERP features also significantly outperformed the PSD features. These results suggest that the tir-aERPs should be potentially more consistent MWL indicators across very different task types when compared to PSD. The current study may provide new insights to our understanding of the common and distinctive neuropsychological essences of MWL across different tasks.

Uncertainty-Guided Mutual Consistency Learning for Semi-Supervised Medical Image Segmentation

Medical image segmentation is a fundamental and critical step in many clinical approaches. Semi-supervised learning has been widely applied to medical image segmentation tasks since it alleviates the heavy burden of acquiring expert-examined annotations and takes the advantage of unlabeled data which is much easier to acquire. Although consistency learning has been proven to be an effective approach by enforcing an invariance of predictions under different distributions, existing approaches cannot make full use of region-level shape constraint and boundary-level distance information from unlabeled data. In this paper, we propose a novel uncertainty-guided mutual consistency learning framework to effectively exploit unlabeled data by integrating intra-task consistency learning from up-to-date predictions for self-ensembling and cross-task consistency learning from task-level regularization to exploit geometric shape information. The framework is guided by the estimated segmentation uncertainty of models to select out relatively certain predictions for consistency learning, so as to effectively exploit more reliable information from unlabeled data. We extensively validate our proposed method on two publicly available benchmark datasets: Left Atrium Segmentation (LA) dataset and Brain Tumor Segmentation (BraTS) dataset. Experimental results demonstrate that our method achieves performance gains by leveraging unlabeled data and outperforms existing semi-supervised segmentation methods.

Evaluating the distinction between semantic knowledge and semantic access: Evidence from semantic dementia and comprehension-impaired stroke aphasia.

Theories of semantic memory based on neuropsychological findings have posited a distinction between stored semantic representations and the mechanisms used to access and manipulate them (e.g., Lambon Ralph, Jefferies, Patterson, & Rogers, 2017; Warrington & Cipolotti, 1996). The most recent instantiation of this view, the controlled semantic cognition theory (Lambon Ralph et al., 2017), is supported by findings suggesting that multimodal (i.e., both verbal and nonverbal) semantic deficits may result from qualitatively different impairments: on the one hand, damage to a semantic access mechanism related to executive control, which is observed in semantic aphasia (SA), and on the other, damage to semantic representations, which is observed in semantic dementia (SD) (Jefferies & Lambon Ralph, 2006). In this study we compared SA and SD patients on several phenomena previously used to support these distinctions. Contrary to the prior results, we found that (1) overall, cross-task consistency was equivalent for the two groups; (2) neither patient group showed consistency driven by item identity across different semantic tasks; (3) correlations among task performance were not obviously driven by the semantic control demands of different tasks; (4) both groups showed executive function deficits; and (5) both groups showed strong effects of distractor interference in a synonym judgment task. Furthermore, we investigated the components of executive ability that could underlie semantic control deficits by correlating performance on updating, shifting, and inhibition tasks with performance on tasks testing semantic abilities. We found that updating was related to semantic processing generally, whereas shifting and inhibition were not. These results also suggest that complex executive function tasks relate to semantic tasks through their shared relationship with language abilities. Overall, evidence from SA and SD patients does not differentiate representations and access mechanisms in the semantic system, as has previously been suggested. Implications for the storage-access distinction are discussed.

The Dialogical Entailment Task

In this paper, a critical discussion is made of the role of entailments in the so-called New Paradigm of psychology of reasoning based on Bayesian models of rationality (Elqayam & Over, 2013). It is argued that assessments of probabilistic coherence cannot stand on their own, but that they need to be integrated with empirical studies of intuitive entailment judgments. This need is motivated not just by the requirements of probability theory itself, but also by a need to enhance the interdisciplinary integration of the psychology of reasoning with formal semantics in linguistics. The constructive goal of the paper is to introduce a new experimental paradigm, called the Dialogical Entailment task, to supplement current trends in the psychology of reasoning towards investigating knowledge-rich, social reasoning under uncertainty (Oaksford & Chater, 2019). As a case study, this experimental paradigm is applied to reasoning with conditionals and negation operators (e.g. CEM and wide and narrow-scope negation). As part of the investigation, participants’ entailment judgments are evaluated against their probability evaluations to assess participants’ cross-task consistency over two experimental sessions.

Consistency, longitudinal stability, and predictions of elementary school students' task interest, success expectancy, and performance in mathematics

This study examined cross-task consistency and longitudinal stability in elementary school students' task interest, success expectancy, and performance from fourth to sixth grade, and their predictive effects on sixth-grade intrinsic value, self-concept, and achievement in mathematics. The results demonstrated consistency in interest, success expectancy, and performance across tasks and stability over time, and these to predict domain-specific motivation and achievement. Virtually no evidence for reciprocal effects was found for task-specific measures, as only previous task performance predicted change in later success expectancy. Cross-lagged effects were observed, however, for predictions of task motivation and performance on domain-specific motivation and achievement, so that success expectancy predicted intrinsic value, interest predicted self-concept, and task performance predicted both self-concept and achievement. Based on the findings, it would seem that students' task-related motivational experiences are associated with their domain-specific beliefs, and that those, in turn, are to some extent manifested in students' task motivation.

Impairments of manual tracking performance during spaceflight are associated with specific effects of microgravity on visuomotor transformations

In contrast to performance in cognitive tasks, tracking performance tends to deteriorate fairly consistently during spaceflight. We address the question whether this decrement results from specific effects of microgravity on motor control or from non-specific effects of the various other stressors present. In a case study we generalize the findings obtained with aiming movements, performed by the same cosmonaut with the same effectors as used for an unstable tracking task, to obtain hypotheses for specific changes of parameters of a simple model used to analyse tracking performance. Consistent with these hypotheses, we observed a reduction of limb stiffness in-flight, but a reduction of the tracking gain post-flight. The cross-task consistency of the observed changes does strongly suggest that the tracking impairment is at least partly caused by specific effects of microgravity on motor control, in particular by a mis-calibration of muscular forces which likely results from an underestimation of masses due to weightlessness.

Cross-task consistency in strategy use and the relationship with intelligence

Two-hundred and forty-three Navy recruits participated in a study that examined the relationship between strategy use and intelligence. Each subject contributed ten enlistment aptitude and achievement scores, took eight additional paper-and-pencil aptitude and intelligence tests, and three computer-administered tests. The computer tests were designed such that as item difficulty increased it became advantageous to strategically shift focus from one aspect of the items to another. Based on reaction time (RT) patterns it was possible to: (a) determine the point along the item difficulty continua where focus shifted, and (b) compare subjects' shift points with the logically optimal shift point. RTs from the computer tests were significantly intercorrelated and were related to psychometric test scores, particularly fluid intelligence. The strategy measures derived from the computer tests were statistically independent from one another and revealed no relationship with any of the psychometric tests. These results cast doubt on previous empirical research and theoretical arguments that strongly link strategy use and general intelligence.

The Emergence and Consolidation of Self-Control from Eighteen to Thirty Months of Age: Normative Trends and Individual Differences

This study is a descriptive report of the capability to exercise self-control in very young children. 2 aspects of self-control were assessed (delay/response inhibition in the presence of an attractive stimulus and compliance with maternal directives in a cleanup task) for 72 children between the ages of 18 and 30 months. The results indicated that both aspects of self-control show age-related increases. However, a factor analysis of the behaviors observed in the cleanup task suggested that compliance could not be adequately described with a unitary, bipolar dimension (noncompliance vs. compliance). 2 patterns of non-compliance were observed, and 1 of these also increased with age. Cross-task consistency for the delay measures) and coherence across the 2 aspects of self-control showed a positive relationship with increasing age. Finally, correlational analyses of the self-control measures and developmental test data showed that individual differences in self-control were associated with differences in cognitive-developmental status (DA). The results are interpreted as evidence that the achievement of self-control can be considered as a major developmental accomplishment and as evidence that individual differences in self-control emerge and are consolidated during the second and third years of life.