Task Model Research Articles

Few-Shot Learning via Repurposing Ensemble of Black-Box Models

This paper investigates the problem of exploiting existing solution models of previous tasks to address a related target task with limited training data. Existing approaches addressing this problem often require access to the internal parameterization of the existing solution models and possibly their training data, which is not possible in many practical settings. To relax this requirement, We approach this problem from a new perspective of black-box re-purposing, which augments the target inputs and leverages their corresponding outputs generated by existing black-box APIs into a feature ensemble. We hypothesize that such feature ensemble can be learned to incorporate and encode relevant black-box knowledge into the feature representation of target data, which will compensate for their scarcity. This hypothesis is confirmed via the reported successes of our proposed black-box ensemble in solving multiple few-shot learning tasks derived from various benchmark datasets. All reported results show consistently that the set of heterogeneous black-box solutions of previous tasks can indeed be reused and combined effectively to solve a reasonably related target task without requiring access to a large training dataset. This is the first step towards enabling new possibilities to further supplement existing techniques in transfer or meta learning with black-box knowledge.

Chinese Spelling Correction as Rephrasing Language Model

This paper studies Chinese Spelling Correction (CSC), which aims to detect and correct potential spelling errors in a given sentence. Current state-of-the-art methods regard CSC as a sequence tagging task and fine-tune BERT-based models on sentence pairs. However, we note a critical flaw in the process of tagging one character to another, that the correction is excessively conditioned on the error. This is opposite from human mindset, where individuals rephrase the complete sentence based on its semantics, rather than solely on the error patterns memorized before. Such a counter-intuitive learning process results in the bottleneck of generalizability and transferability of machine spelling correction. To address this, we propose Rephrasing Language Modeling (ReLM), where the model is trained to rephrase the entire sentence by infilling additional slots, instead of character-to-character tagging. This novel training paradigm achieves the new state-of-theart results across fine-tuned and zero-shot CSC benchmarks, outperforming previous counterparts by a large margin. Our method also learns transferable language representation when CSC is jointly trained with other tasks.

Generalizable Task Representation Learning for Offline Meta-Reinforcement Learning with Data Limitations

Generalization and sample efficiency have been long-standing issues concerning reinforcement learning, and thus the field of Offline Meta-Reinforcement Learning (OMRL) has gained increasing attention due to its potential of solving a wide range of problems with static and limited offline data. Existing OMRL methods often assume sufficient training tasks and data coverage to apply contrastive learning to extract task representations. However, such assumptions are not applicable in several real-world applications and thus undermine the generalization ability of the representations. In this paper, we consider OMRL with two types of data limitations: limited training tasks and limited behavior diversity and propose a novel algorithm called GENTLE for learning generalizable task representations in the face of data limitations. GENTLE employs Task Auto-Encoder (TAE), which is an encoder-decoder architecture to extract the characteristics of the tasks. Unlike existing methods, TAE is optimized solely by reconstruction of the state transition and reward, which captures the generative structure of the task models and produces generalizable representations when training tasks are limited. To alleviate the effect of limited behavior diversity, we consistently construct pseudo-transitions to align the data distribution used to train TAE with the data distribution encountered during testing. Empirically, GENTLE significantly outperforms existing OMRL methods on both in-distribution tasks and out-of-distribution tasks across both the given-context protocol and the one-shot protocol.

Discretization-Induced Dirichlet Posterior for Robust Uncertainty Quantification on Regression

Uncertainty quantification is critical for deploying deep neural networks (DNNs) in real-world applications. An Auxiliary Uncertainty Estimator (AuxUE) is one of the most effective means to estimate the uncertainty of the main task prediction without modifying the main task model. To be considered robust, an AuxUE must be capable of maintaining its performance and triggering higher uncertainties while encountering Out-of-Distribution (OOD) inputs, i.e., to provide robust aleatoric and epistemic uncertainty. However, for vision regression tasks, current AuxUE designs are mainly adopted for aleatoric uncertainty estimates, and AuxUE robustness has not been explored. In this work, we propose a generalized AuxUE scheme for more robust uncertainty quantification on regression tasks. Concretely, to achieve a more robust aleatoric uncertainty estimation, different distribution assumptions are considered for heteroscedastic noise, and Laplace distribution is finally chosen to approximate the prediction error. For epistemic uncertainty, we propose a novel solution named Discretization-Induced Dirichlet pOsterior (DIDO), which models the Dirichlet posterior on the discretized prediction error. Extensive experiments on age estimation, monocular depth estimation, and super-resolution tasks show that our proposed method can provide robust uncertainty estimates in the face of noisy inputs and that it can be scalable to both image-level and pixel-wise tasks.

Greater Risk Taking in Cosmetic Surgery Acceptance and History: An Experimental and Computational Study

Cosmetic surgery is ever more affordable and accessible, but carries physical and psychological risks. Yet, no study to date has directly examined risk-taking behaviour under controlled conditions, beyond self-report and in relation to cosmetic surgery attitudes. We used the Balloon Analogue Risk Task and advanced computational modelling to measure decision-making behaviour and identify the latent parameters driving behaviour associated with cosmetic surgery attitudes in women with no cosmetic surgery history (N = 265) and a subsample of women with a cosmetic surgery history (N = 24). Risk taking was higher in women with greater acceptance and history of cosmetic surgery. Computational modelling revealed increased risk taking in women with greater acceptance of cosmetic surgery when decisions were made with greater knowledge of loss (risk) and not when the likelihood of loss was unknown (uncertainty). When women with greater acceptance of cosmetic surgery made decisions, they also placed less emphasis on possible losses (reduced loss aversion). Our findings suggest that women seeking cosmetic procedures may be less sensitive to losses and thus make more risky decisions. Greater emphasis should be placed on communicating potential losses rather than just the associated risks to women considering cosmetic procedures.No Level Assigned This journal requires that authors assign a level of evidence to each submission to which Evidence-Based Medicine rankings are applicable. This excludes Review Articles, Book Reviews, and manuscripts that concern Basic Science, Animal Studies, Cadaver Studies, and Experimental Studies. For a full description of these Evidence-Based Medicine ratings, please refer to the Table of Contents or the online Instructions to Authors www.springer.com/00266.

A proactive energy management strategy for battery-powered autonomous systems

Battery energy management systems have been studied in control communities for many years. This paper proposes a new perspective by integrating control and scheduling for battery-powered autonomous systems. This is motivated by the observations that battery closed-loop control can significantly improve the DC-bus stability but reduce the battery durability, while load scheduling can considerably improve the battery durability by smoothing the load power. In view of the above findings, we propose a proactive energy management strategy for the battery energy management system to improve both the DC-bus stability and battery durability. We first analyze the schedulability of load tasks to check if they are schedulable. Then, the power of schedulable loads is scheduled with an active load scheduling algorithm to smooth the fluctuation of battery current and extend the battery lifetime. Thereafter, the scheduled load power is integrated with a feedforward control to restrain DC-bus voltage fluctuation. In addition to the classical sporadic/periodic load tasks model, we propose a new aperiodic load task model to characterize the load power triggered by events in practical applications. An experimental platform is built to verify the effectiveness of the proposed proactive energy management method. Experimental results show that the proposed proactive energy management method can suppress the 15.71% DC-bus voltage fluctuation and reduce the 8.93% battery current fluctuation, extending the battery lifetime by 7.57% compared to existing energy management strategies.

Air–Ground Collaborative Multi-Target Detection Task Assignment and Path Planning Optimization

Collaborative exploration in environments involving multiple unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs) represents a crucial research direction in multi-agent systems. However, there is still a lack of research in the areas of multi-target detection task assignment and swarm path planning, both of which play a vital role in enhancing the efficiency of environment exploration and reducing energy consumption. In this paper, we propose an air–ground collaborative multi-target detection task model based on Mixed Integer Linear Programming (MILP). In order to make the model more suitable for real situations, kinematic constraints of the UAVs and UGVs, dynamic collision avoidance constraints, task allocation constraints, and obstacle avoidance constraints are added to the model. We also establish an objective function that comprehensively considers time consumption, energy consumption, and trajectory smoothness to improve the authenticity of the model and achieve a more realistic purpose. Meanwhile, a Branch-and-Bound method combined with the Improved Genetic Algorithm (IGA-B&B) is proposed to solve the objective function, and the optimal task assignment and optimal path of air–ground collaborative multi-target detection can be obtained. A simulation environment with multi-agents, multi-obstacles, and multi-task points is established. The simulation results show that the proposed IGA-B&B algorithm can reduce the computation time cost by 30% compared to the traditional Branch-and-Bound (B&B) method. In addition, an experiment is carried out in an outdoor environment, which further validates the effectiveness and feasibility of the proposed method.

A guide to measuring expert performance in forensic pattern matching

Decisions in forensic science are often binary. A firearms expert must decide whether a bullet was fired from a particular gun or not. A face comparison expert must decide whether a photograph matches a suspect or not. A fingerprint examiner must decide whether a crime scene fingerprint belongs to a suspect or not. Researchers who study these decisions have therefore quantified expert performance using measurement models derived largely from signal detection theory. Here we demonstrate that the design and measurement choices researchers make can have a dramatic effect on the conclusions drawn about the performance of forensic examiners. We introduce several performance models – proportion correct, diagnosticity ratio, and parametric and non-parametric signal detection measures – and apply them to forensic decisions. We use data from expert and novice fingerprint comparison decisions along with a resampling method to demonstrate how experimental results can change as a function of the task, case materials, and measurement model chosen. We also graphically show how response bias, prevalence, inconclusive responses, floor and ceiling effects, case sampling, and number of trials might affect one’s interpretation of expert performance in forensics. Finally, we discuss several considerations for experimental and diagnostic accuracy studies: (1) include an equal number of same-source and different-source trials; (2) record inconclusive responses separately from forced choices; (3) include a control comparison group; (4) counterbalance or randomly sample trials for each participant; and (5) present as many trials to participants as is practical.

Different Dual-Task Paradigm Reduce Postural Control Ability and Dynamic Stability of Healthy Young Adults during Stair Descent.

This study aimed to compare the impacts of different dual-task paradigms on the postural control ability and dynamic stability of the youth during stair descent. Twenty young adults without regular exercise habits were randomly recruited to perform stair descent tasks with three different paradigms: single-task, cognitive dual-task, and manual dual-task. Kinematic and dynamic data were collected using an 8 Vicon motion analysis system and a Kistler force plate to evaluate postural control ability and dynamic stability during stair descent. The variation trends of lower limb joint moment were similar under the three task models. Compared with a single-task, both dual-task paradigms significantly reduced the mechanical parameters and dynamic stability during stair descent. The dual-task paradigm increases the risk of stair-related falls. Both cognitive and manual tasks have similar impacts on postural control ability and dynamic stability during stair walking. It is recommended that people avoid performing dual tasks during stair descent.

Drop the shortcuts: image augmentation improves fairness and decreases AI detection of race and other demographics from medical images

It has been shown that AI models can learn race on medical images, leading to algorithmic bias. Our aim in this study was to enhance the fairness of medical image models by eliminating bias related to race, age, and sex. We hypothesise models may be learning demographics via shortcut learning and combat this using image augmentation. This study included 44,953 patients who identified as Asian, Black, or White (mean age, 60.68 years±18.21; 23,499 women) for a total of 194,359 chest X-rays (CXRs) from MIMIC-CXR database. The included CheXpert images comprised 45,095 patients (mean age 63.10 years±18.14; 20,437 women) for a total of 134,300 CXRs were used for external validation. We also collected 1195 3D brain magnetic resonance imaging (MRI) data from the ADNI database, which included 273 participants with an average age of 76.97 years±14.22, and 142 females. DL models were trained on either non-augmented or augmented images and assessed using disparity metrics. The features learned by the models were analysed using task transfer experiments and model visualisation techniques. In the detection of radiological findings, training a model using augmented CXR images was shown to reduce disparities in error rate among racial groups (-5.45%), age groups (-13.94%), and sex (-22.22%). For AD detection, the model trained with augmented MRI images was shown 53.11% and 31.01% reduction of disparities in error rate among age and sex groups, respectively. Image augmentation led to a reduction in the model's ability to identify demographic attributes and resulted in the model trained for clinical purposes incorporating fewer demographic features. The model trained using the augmented images was less likely to be influenced by demographic information in detecting image labels. These results demonstrate that the proposed augmentation scheme could enhance the fairness of interpretations by DL models when dealing with data from patients with different demographic backgrounds. National Science and Technology Council (Taiwan), National Institutes of Health.

"Machine replacement" or "job creation": How does artificial intelligence impact employment patterns in China's manufacturing industry?

Artificial intelligence (AI), as an important engine for promoting high-quality economic development, should not be overlooked in terms of its impact on the employment of the labor force while promoting the digital and intelligent transformation of industries. In the face of the complex international environment and non-systemic shocks, it is of great significance to explore whether it is "machine replacement" or "job creation" in the process of the integration of AI and industry, as well as the impact of technological progress on the employment pattern of the labor force, in order to promote the economic development, respond to and solve the employment problem. It is of great significance to promote economic development and cope with and solve the employment problem. Based on the task model, this paper analyses the mechanism of the impact of AI on the employment pattern of manufacturing industry. Meanwhile, based on the provincial panel data of China's manufacturing industry from 2011 to 2020, it empirically examines the impact of AI on the total employment, employment structure and employment quality of the labor force, and analyses the multiple responses of AI on the employment pattern of the manufacturing industry. The study shows that: Firstly, the level of development of AI and the total amount of employment is a positive U-shaped relationship, the short term is dominated by the substitution effect, and the long term is dominated by the creation effect; Secondly, with regard to the employment structure, low-skilled labor is more likely to be replaced. The financial, accommodation and catering industries are relatively less affected by the spillover effects of the manufacturing industry; Third, with regard to the employment quality, the gap between urban and rural incomes has eased, with per capita net income of rural residents rising to a higher degree than per capita disposable income of urban residents. Thus, in order to further address the impact of AI on the employment patterns of the labor force, the level of AI development should be increased while expanding employment channels, paying attention to labor force skills training, reinforcing the leading role of developed regions, and accelerating regional integration and urban-rural integration, so as to share the dividends of technological progress.

Modeling and Control Strategies for Flight Service Vehicle Queue Behavior in Non Cooperative Games

From the perspective of the running process for service vehicles, based on the analysis of the benefits maximization of the special needs vehicles joining the queue as the scheduling goal, and aiming at the shortcomings of traditional task scheduling strategies, methods and models, a non-cooperative game behavior optimization scheduling model with security, economy and environmental protection as the optimization goals was proposed by using game theory. At the same time, the vehicle service capability, service times and import Angle are limited to ensure that the model is more in line with the actual operation scenario. In order to judge the deviation between the game result and the actual operation, the game efficiency matrix is constructed, and Nash equilibrium is used to prove that the model has an equilibrium point. On the basis of the game model, the strategy layer, the game layer and the optimization layer are divided into queue behavior control strategy methods. Finally, this paper verifies the feasibility of the model by using the actual parameter changes to determine the best parameter combination suitable for the model.

State-of-Health Estimation of Li-Ion Batteries Using Semiparametric Adaptive Transfer Learning

For lithium-ion batteries (LIBs), precise state of health (SoH) estimation can provide direction for reasonable use, minimize battery failure rates, and extend battery life. Data-driven methods are promising for SoH estimation since they work effectively without requiring human interaction and have great nonlinear prediction capabilities. The majority of studies also found that the training data was sufficient. However, in real-world contexts, data collection is frequently time-consuming and expensive. This paper suggests using transfer learning to make the model less dependent on data for tracking battery state of health (SoH). Many approaches assume training and testing data have the same distribution. Due to distribution mismatch, a model that works for one data set may not for another. We propose an adaptive transfer learning technique relying on Gaussian Processes regression (AT-GPR) in this research. Automatically evaluating source-to-target task similarity can construct learning systems. Bayesian semi-parametric transfer kernel proposed to learn target task model. Battery cyclic aging data from two publicly available data sets in various working environments are considered to validate the suggested method. The experiments show that AT-GPR generates reliable prediction results; however, only 20% of the overall data set is made up of training data.

The time course of the spatial representation of ‘past’ and ‘future’ concepts: New evidence from the STEARC effect

Humans use space to think of and communicate the flow of time. This spatial representation of time is influenced by cultural habits so that in left-to-right reading cultures, short durations and past events are mentally positioned to the left of long durations and future events. The STEARC effect (Space Temporal Association of Response Codes) shows a faster classification of short durations/past events with responses on the left side of space and of long durations/future events with responses on the right side. We have recently showed that during the classification of time durations, space is a late heuristic of time because in this case, the STEARC appears only when manual responses are slow, not when they are fast. Here, we wished to extend this observation to the semantic classification of words as referring to the ‘past’ or the ‘future’. We hypothesised that the semantic processing of ‘past’ and ‘future’ concepts would have engaged slower decision processes than the classification of short versus long time durations. According to dual-route models of conflict tasks, if the task-dependent classification/decision process were to proceed relatively slowly, then the effects of direct activation of culturally preferred links between stimulus and response (S-R), i.e., past/left and future/right in the case of the present task, should attain higher amplitudes before the instruction-dependent correct response is selected. This would imply that, at variance with the faster classification of time durations, during the slower semantic classification of time concepts, in incongruent trials, the direct activation of culturally preferred S-R links should introduce significant reaction time (RT) costs and a corresponding STEARC at the fastest manual responses in the experiment too. The study's results confirmed this hypothesis and showed that in the classification of temporal words, the STEARC also increased as a function of the length of RTs. Taken together, the results from sensory duration and semantic classification STEARC tasks show that the occurrence, strength and time course of the STEARC varies significantly as a function of the speed and level of cognitive processing required in the task.

Teachers' perceptions of the epistemic aims and evaluation criteria of multiple text integration

The importance of intertextual integration in current information societies has led to extensive research on how to foster students' integration competence. Although teachers have a central role in promoting integration skills, research into their perceptions of intertextual integration tasks is still scarce. Hence, the purpose of this study was to explore language arts teachers' views on the epistemic aims and evaluation criteria of intertextual integration. Fifty-four language arts teachers were asked to evaluate three integrated essays, supposedly written by students, that reflected typical violations of key integration quality criteria. In addition, they responded to open-ended questions about intertextual integration aims and evaluation criteria. Teachers successfully identified only about a third of the integration quality criteria violations that were embedded in the essays. Nonetheless, they collectively raised diverse evaluation criteria that addressed content selection and understanding, argumentation quality, source use and intertextuality, structure and organization, and language sophistication. Teachers' most prevalent aims were related to information processing and to the development of learners’ reading and writing skills. However, teachers also referred to knowledge-building goals. Quantitative and qualitative analyses revealed interrelations between teachers' perceptions of evaluation criteria and their integration aims. Overall, these findings demonstrate that language arts teachers have rich multiple-text integration task models. At the same time, they suggest that many teachers might not perceive intertextual integration as an opportunity for knowledge building and for developing learners’ critical thinking. The results of this study can inform professional development efforts to support instruction of intertextual integration.

Regional choroidal thickness estimation from color fundus images based on convolutional neural networks

PurposeThis study aims to estimate the regional choroidal thickness from color fundus images from convolutional neural networks in different network structures and task learning models. Method1276 color fundus photos and their corresponding choroidal thickness values from healthy subjects were obtained from the Topcon DRI Triton optical coherence tomography machine. Initially, ten commonly used convolutional neural networks were deployed to identify the most accurate model, which was subsequently selected for further training. This selected model was then employed in combination with single-, multiple-, and auxiliary-task training models to predict the average and sub-region choroidal thickness in both ETDRS (Early Treatment Diabetic Retinopathy Study) grids and 100-grid subregions. The values of mean absolute error and coefficient of determination (R2) were involved to evaluate the models' performance. ResultsEfficientnet-b0 network outperformed other networks with the lowest mean absolute error value (25.61 μm) and highest R2 (0.7817) in average choroidal thickness. Incorporating diopter spherical, anterior chamber depth, and lens thickness as auxiliary tasks improved predicted accuracy (p-value = 6.39×10−44, 2.72×10−38, 1.15×10−36 respectively). For ETDRS regional choroidal thickness estimation, multi-task model achieved better results than single task model (lowest mean absolute error = 31.10 μm vs. 33.20 μm). The multi-task training also can simultaneously predict the choroidal thickness of 100 grids with a minimum mean absolute error of 33.86 μm. ConclusionsEfficientnet-b0, in combination with multi-task and auxiliary task models, achieve high accuracy in estimating average and regional macular choroidal thickness directly from color fundus photographs.

A Methodology for Ship Cabin Equipment Layout Considering Human Factor Reliability Optimization

_ The paper proposes an innovative methodology of Human Cognitive Reliability- Cognitive Reliability and Error Analysis Method (HCR-CREAM) coupled with A* search and genetic algorithm (GA) to tackle ship cabin equipment layout considering human factor reliability optimization with the goal of minimizing human error probability (HEP) subjected to practical requirements. After establishing the mathematical model of cabin equipment inspection tasks in ship cabin equipment layout problem through HCR-CREAM and equipment geometric simplification, a method of the horizontal movement based on minimum distance is presented to avoid the equipment overlapping, then A* search is used for planning inspection paths and GA with selection, crossover, and mutation operators is applied to solve equipment layout results. A case of equipment layout in a certain ship engine room has been taken to carry out parameter sampling experiments by Latin Hypercube for GA. The results show the solution effect of GA is less affected by its parameter variation. And through the comparison with the initial equipment layout, the indicators influencing the HEP of the optimized result have been improved, thus significantly reducing HEP. Keywords ship cabin equipment layout; human factor reliability optimization; HCRCREAM; A* search; genetic algorithm

Моделирование учебных планов профильного обучения

The article focuses on the need to modernize approaches to organizing specialized education in high school in connection with the change in vectors of economic and technological development of the country, which are reflected in the latest decisions of the President and government of the Russian Federation. To achieve this task, an organizational and content model of specialized training is presented, which makes it possible to combine the goals of participants in educational relations, including the state, and the requirements of the regulatory framework of Russian education. The result of this combination is a curriculum for specialized training, examples of the compilation of which for a technological profile of an engineering orientation complete the article.

Machine Learning Technique Reveals Intrinsic EEG Connectivity Characteristics of Patients With Mild Stroke During Cognitive Task Performing

Although convergent evidence has shown that patients with mild stroke (MS) are commonly accompanied by post-stroke cognitive and/or memory impairment, only disproportionate attention was paid compared to severe stroke. To promote post-stroke management for early intervention in MS-related cognitive impairment, a feasible and convenient method for MS detection is therefore favorable. A data-driven classification framework combined with quantitative graph theoretical analysis was introduced in this work, aiming to provide a comprehensive appreciation of MS-related brain network alterations. EEG functional connectivity (FC) was constructed from 45 patients with MS and 45 healthy participants during two cognitive tasks (i.e., visual and auditory oddball) and set as input for the classification model and graph theoretical analysis. As expected, patients showed significantly reduced behavioral performance in both tasks. Furthermore, we achieved a satisfactory classification accuracy of 88.9% with a decision fusion strategy from classification models of both tasks. The spatio-spectral characteristics of the discriminative FC revealed complex topological distributions in both tasks. Moreover, significantly decreased global efficiency was found, suggesting a MS-related disruption in parallel information processing. Overall, these results demonstrated the potential of FC as salient biomarkers for detecting MS, and extended our understanding of the underlying MS-related neural mechanisms during cognitive processing.

Task-Based Compliance Control for Bottle Screw Manipulation With a Dual-Arm Robot

In this paper, a novel task-based compliance control approach for a dual-arm robot is presented with a bottle screw task. The presented approach aims at overcoming uncertainties from the object model and contact forces during the bottle screw task. A novel framework is proposed to synthesize the task motion planning and compliance control that ensure desired performance of both accuracy and compliant motion. The proposed task-based compliance control approach provides a hierarchical strategy: gross motion planning and fine compliance motion planning. The gross motion planning involves the absolute and relative motion control on a macro scale, while the fine compliance motion planning deals with uncertainties by the compliance control to accomplish a task requiring high precision robustly. A theoretical modeling of the bottle screw task is presented within the proposed framework through the analysis of uncertainties and constraints. The experimental results show that the proposed framework is efficient and robust to operate a set of bottles.