Relevant Subtasks Research Articles

Human-AI Collaborative Sub-Goal Optimization in Hierarchical Reinforcement Learning

Hierarchical reinforcement learning often involves human expertise in defining multiple sub-goals to decompose complex objectives into relevant sub-tasks. However, manually specifying these sub-goals is labor-intensive, costly, and prone to introducing biases or misleading the agent. To overcome these challenges, we propose a collaborative human-AI algorithm that seamlessly integrates with hierarchical models to automatically update prior knowledge and optimize candidate sub-goals. Our algorithm can be easily incorporated into a wide range of goal-conditioned frameworks. We evaluate our approach in comparison with relevant baselines, we demonstrate the effectiveness of our algorithm in addressing and preventing negative inferences arising from confusing or conflicting sub-goals. Additionally, our algorithm shows robustness across different levels of human knowledge, accelerating convergence towards optimal sub-goal spaces and hierarchical policies.

Virtual reality assessment of functional capacity in people with Schizophrenia: Associations with reduced emotional experience and prediction of functional outcomes

Virtual Reality (VR) approaches have had considerable success in measurement of functional capacity. However, it is not clear if factors other than cognitive impairment influence performance on VR measures. Many people with schizophrenia have significant negative symptoms and they could reduce engagement in assessment. 158 patients with schizophrenia performed the VRFCAT, were tested with the MCCB, were rated with the PANSS, and were rated on everyday functioning. Scores for reduced emotional experience and reduced expression were derived. Reduced emotional experience, but not reduced expression, was correlated with socially relevant VRFCAT subtasks and real-world social functioning. Performance on the socially relevant subtasks, but not the solitary subtasks, shared variance with work outcomes. MCCB performance was associated with both subdomains, but socially relevant subtasks shared more variance. Patients with higher reduced emotional experience validly engaged in socially relevant VR simulations, as indexed by correlations with outcome measures. These patients had poorer performance on socially relevant tasks than on solitary tasks. The differential validity of solitary vs. socially relevant simulations was supported by differences in correlates, suggesting that assessments with a focus on performance of simulated socially relevant tasks could be developed.

An ontology-based exploration of the concepts and relationships in the activities and participation component of the international classification of functioning, disability and health

BackgroundThe International Classification of Functioning, Disability and Health (ICF) is a classification of health and health-related issues, aimed at describing and measuring health and disability at both individual and population levels. Here we discuss a preliminary qualitative and quantitative analysis of the relationships used in the Activities and Participation component of ICF, and a preliminary mapping to SUMO (Suggested Upper Merged Ontology) concepts. The aim of the analysis is to identify potential logical problems within this component of ICF, and to understand whether activities and participation might be defined more formally than in the current version of ICF.ResultsIn the relationship analysis, we used four predicates among those available in SUMO for processes (Patient, Instrument, Agent, and subProcess). While at the top level subsumption was used in most cases (90%), at the lower levels the percentage of other relationships rose to 41%. Chapters were heterogeneous in the relationships used and some of the leaves of the tree seemed to represent properties or parts of the parent concept rather than subclasses. Mapping of ICF to SUMO proved partially feasible, with the activity concepts being mapped mostly (but not totally) under the IntentionalProcess concept in SUMO. On the other hand, the participation concept has not been mapped to any upper level concept.ConclusionsOur analysis of the relationships within ICF revealed issues related to confusion between classes and their properties, incorrect classifications, and overemphasis on subsumption, confirming what already observed by other researchers. However, it also suggested some properties for Activities that could be included in a more formal model: number of agents involved, the instrument used to carry out the activity, the object of the activity, complexity of the task, and an enumeration of relevant subtasks.

An ontology-based exploration of the concepts and relationships in the Activities and Participation component of the International Classification of Functioning, Disability and Health

The International Classification of Functioning, Disability and Health (ICF) is a classification of health and health-related issues, aimed at describing and measuring health and disability at both individual and population levels. Here we discuss a preliminary qualitative and quantitative analysis of the relationships used in the Activities and Participation component of ICF, and a preliminary mapping to SUMO (Suggested Upper Merged Ontology) concepts. The aim of the analysis is to identify potential logical problems within this component of ICF, and to understand whether activities and participation might be defined more formally than in the current version of ICF. In the relationship analysis, we used four predicates among those available in SUMO for processes (Patient, Instrument, Agent, and subProcess). While at the top level subsumption was used in most cases (90%), at the lower levels the percentage of other relationships rose to 41%. Chapters were heterogeneous in the relationships used and some of the leaves of the tree seemed to represent properties or parts of the parent concept rather than subclasses. Mapping of ICF to SUMO proved partially feasible, with the activity concepts being mapped mostly (but not totally) under the IntentionalProcess concept in SUMO. On the other hand, the participation concept has not been mapped to any upper level concept. Our analysis of the relationships within ICF revealed issues related to confusion between classes and their properties, incorrect classifications, and overemphasis on subsumption, confirming what already observed by other researchers. However, it also suggested some properties for Activities that could be included in a more formal model: number of agents involved, the instrument used to carry out the activity, the object of the activity, complexity of the task, and an enumeration of relevant subtasks.

Probabilistic joint models incorporating logic and learning via structured variational approximation for information extraction

Traditional information extraction systems for compound tasks adopt pipeline architectures, which are highly ineffective and suffer from several problems such as cascading accumulation of errors. In this paper, we propose a joint discriminative probabilistic framework to optimize all relevant subtasks simultaneously. This framework offers a great flexibility to incorporate the advantage of both uncertainty for sequence modeling and first-order logic for domain knowledge. The first-order logic model provides a more expressive formalism tackling the issue of limited expressiveness of traditional attribute-value representation. Our framework defines a joint probability distribution for both segmentations in sequence data and possible worlds of relations between segments in the form of an exponential family. Since exact parameter estimation and inference are prohibitively intractable in this model, a structured variational inference algorithm is developed to perform parameter estimation approximately. For inference, we propose a highly coupled, bi-directional Metropolis-Hastings (MH) algorithm to find the maximum a posteriori (MAP) assignments for both segmentations and relations. Extensive experiments on two real-world information extraction tasks, entity identification and relation extraction from Wikipedia, and citation matching show that (1) the proposed model achieves significant improvement on both tasks compared to state-of-the-art pipeline models and other joint models; (2) the bi-directional MH inference algorithm obtains boosted performance compared to the greedy, N-best list, and uni-directional MH sampling algorithms.

Task-Oriented Modular Decomposition of Biological Networks: Trigger Mechanism in Blood Coagulation

Analysis of complex time-dependent biological networks is an important challenge in the current postgenomic era. We propose a middle-out approach for decomposition and analysis of complex time-dependent biological networks based on: 1), creation of a detailed mechanism-driven mathematical model of the network; 2), network response decomposition into several physiologically relevant subtasks; and 3), subsequent decomposition of the model, with the help of task-oriented necessity and sensitivity analysis into several modules that each control a single specific subtask, which is followed by further simplification employing temporal hierarchy reduction. The technique is tested and illustrated by studying blood coagulation. Five subtasks (threshold, triggering, control by blood flow velocity, spatial propagation, and localization), together with responsible modules, can be identified for the coagulation network. We show that the task of coagulation triggering is completely regulated by a two-step pathway containing a single positive feedback of factor V activation by thrombin. These theoretical predictions are experimentally confirmed by studies of fibrin generation in normal, factor V-, and factor VIII-deficient plasmas. The function of the factor V-dependent feedback is to minimize temporal and parametrical intervals of fibrin clot instability. We speculate that this pathway serves to lessen possibility of fibrin clot disruption by flow and subsequent thromboembolism.