Multi-task Neural Network Research Articles

A novel method for approaching the compatibility of tree biomass estimation by multi-task neural networks

It is important to guarantee the property of biological compatibility when estimating tree biomass of the total and components for carbon accounting under global climate change. The issue was successfully considered in traditional nonlinear regression models, but not for machine learning methods. A new method for approaching the compatibility of tree biomass estimation in ANN (Artificial Neural Network) was developed by using the multi-task loss function, which had the desire features of minimizing residuals and approaching biomass compatibility. The method was tested by two tree species biomass dataset and showed the desired feature. Leave-one-out validation results showed that comparing ANN model with simultaneously fitting 7 outputs (stem, bark, branch, leaf, crown, trunk, aboveground) and classical loss function, the RMSE of aboveground estimation (AGB) and the mean absolute relative difference between AGB and the sum of component biomass estimations from the model developed by our new method decreased from 166.864 (kg) to 154.860 (kg) and from 4.757% to 0.071%, respectively for Abies nephrolepis dataset, and from 49.18 (kg) to 33.060 (kg) and from 5.314% to 0.636%, respectively for Acer mono dataset. It provided a trade-off solution for the error accumulation and the compatibility among components and the total estimations when using ANN for tree biomass modelling, and was useful for carbon accounting using machine learning methods.

A Multiobjective Cooperative Driving Framework Based on Evolutionary Algorithm and Multitask Learning

The development of connected and automated vehicle (CAV) techniques brings an upcoming revolution to traffic management. The control of CAVs in potential conflict areas such as on-ramps and intersections will be complex to traffic management when considering their deployment. There is still a lack of a general framework for dispatching CAVs in these bottlenecks, which is expected to ensure safety, traffic efficiency, and energy consumption in real time. This study aimed to fill the technique gap, and a comprehensive cooperative intelligent driving framework is put forward to study the problem, which can be used in both on-ramp and intersection scenarios. Based on a multi-objective evolutionary algorithm, CAVs are denoted as a sequence to be searched in solution space, while a multitask learning neural network with adaptive loss function is implemented for optimization target feedback to surrogate the simulation test procedure. The simulation results show that the proposed framework can get satisfying performance with low time and energy consumption. It can reduce time consumption by up to 16.51% for the on-ramp scenario and 9.8% for the intersection scenario, while reducing energy consumption by up to 16.39% and 11.39% for the two scenarios. Meanwhile, an analysis of computation time is carried out, illuminating the flexibility and controllability of the new strategy.

Topic Break Detection in Interview Dialogues Using Sentence Embedding of Utterance and Speech Intention Based on Multitask Neural Networks.

Currently, task-oriented dialogue systems that perform specific tasks based on dialogue are widely used. Moreover, research and development of non-task-oriented dialogue systems are also actively conducted. One of the problems with these systems is that it is difficult to switch topics naturally. In this study, we focus on interview dialogue systems. In an interview dialogue, the dialogue system can take the initiative as an interviewer. The main task of an interview dialogue system is to obtain information about the interviewee via dialogue and to assist this individual in understanding his or her personality and strengths. In order to accomplish this task, the system needs to be flexible and appropriate for detecting topic switching and topic breaks. Given that topic switching tends to be more ambiguous in interview dialogues than in task-oriented dialogues, existing topic modeling methods that determine topic breaks based only on relationships and similarities between words are likely to fail. In this study, we propose a method for detecting topic breaks in dialogue to achieve flexible topic switching in interview dialogue systems. The proposed method is based on multi-task learning neural network that uses embedded representations of sentences to understand the context of the text and utilizes the intention of an utterance as a feature. In multi-task learning, not only topic breaks but also the intention associated with the utterance and the speaker are targets of prediction. The results of our evaluation experiments show that using utterance intentions as features improves the accuracy of topic separation estimation compared to the baseline model.

Dynamic feeding method for aquaculture fish using multi-task neural network

In recirculating aquaculture system (RAS), fish feeding is the most important part in production management, which is not only related to economic benefits, but also the key to ensure fish welfare and increase production. At present, in RAS, fish are basically fed either artificially or automatically (quantitatively supply feed at definite time), which can easily result in under-feeding or over-feeding of fish. Therefore, there is an urgent to develop an intelligent method that realizes appropriate feeding according to the actual demands of fish. This research attempts to explore a fish dynamic feeding method based on the multi-task network to meet the automatic adjustment of both the feeding intervals (the time intervals between feeding points in repeated feeding in a single-round) and feeding rates. The specific objectives of this study include two parts: 1) to construct a multi-task network to analyze the feeding activity of cultured fish and monitor the amount of uneaten feed pellets; 2) to design a feeding strategy based on information obtained from the multi-task network that realizes the dynamic adjustment of feeding intervals and the decision of feeding endpoint. The waste of feed pellets can be reduced by dynamically adjusting the feeding intervals, and the under-feeding and over-feeding of fish can be prevented by determining feeding endpoint. The results indicated that the accuracy of feeding activity classification by multi-task network reached 95.44%, and the mean absolute error (MAE) and mean square error (MSE) in uneaten feed pellet counting were 4.80 and 6.75, which indicate that the multi-task network can accurately monitor the fish feeding activity and the amount of uneaten feed pellets. Based on the two monitored information, combined with the feeding strategy, we dynamically adjusted the feeding intervals and determined the feeding endpoint, and then compared the feeding endpoints with manual judgment to verify the feasibility and accuracy of the dynamic feeding method based on the multi-task network. In summary, this research provides a more accurate and efficient solution for the intelligent and precise feeding of cultured fish, and provides the theoretical foundation for the development of intelligent feeding devices.

Detection Scheme With Joint Intertrack Interference and Media Noise Mitigations for Heat Assisted Interlaced Magnetic Recording

Heat assisted interlaced magnetic recording (HIMR) is a promising candidate for the next-generation of magnetic recording technology to further increase the area density beyond 1Tb/in <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> . Specifically, the high temperature and low temperature tracks are written in an interlaced order to improve the recording performance. However, the inter-track interference (ITI), inter-symbol interference (ISI) and thermal jitters brought by the increased recording density and Curie temperature variations are severe in HIMR, which degrade the bit error rate (BER) performance obviously. In this study, we propose a multitrack detection scheme with joint intertrack interference and media noise mitigations. Here a multi-task neural network (MTNN) is designed to simultaneously predict ITI pattern and residual media noise, then the 2D variable equalizers corresponding to different ITI patterns are implemented and predicted residual media noises are embedded into the branch metrics of modified Bahl-Cocke-Jelinek-Raviv (BCJR) detector to mitigate ITI and whiten media noise. The simulation demonstrates that the proposed MTNN with variable equalizer and modified BCJR detector (MTNN+VE+MB) algorithm mitigates the ITI and media noise effectively. At the channel bit density of 3.10 Tb/in <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> , it provides 2.6 dB signal-to-noise ratio (SNR) gain compared to that of conventional 2D fixed equalizer with pattern dependent noise prediction detector (FE+PDNP) for the low temperature (LT) tracks with 4% Curie temperature variance.

Propensity Score Analysis with Missing Data Using a Multi-Task Neural Networks

Propensity Score Analysis with Missing Data Using a Multi-Task Neural Networks

Multitask Deep Learning for Segmentation and Lumbosacral Spine Inspection

Multi-task learning has achieved notable progress in many medical applications. In this paper, we propose a multi-task neural network, MRNet, for segmentation and spinal parameter inspection. It is developed based on the multi-path convolutional neural network for the robust detection of obscure regions on X-ray images. The proposed MRNet has two branches. One is for the segmentation of lumbar vertebrae, sacrum, and femoral heads. It shares the main features with the second branch for detection and classification by supervised learning. The output of the second branch is used to estimate the parameters for lumbosacral spine inspection. We conduct this research on our dataset collected and annotated by doctors for model training and performance evaluation. The datasets are used to train our MRNet as well as other networks for performance evaluation. Compared to the state-of-the-art techniques, the proposed MRNet is capable of X-ray image segmentation and parameter estimation with very limited training data. The results have demonstrated the feasibility of our MRNet for the segmentation of lumbar vertebrae, as well as the automated parameter prediction for lumbosacral spine inspection. Code is available at https://github.com/LuanTran07/BiLUnet-Lumbar-Spine.

A Multi-Task Network Approach for Calculating Discrimination-Free Insurance Prices

In applications of predictive modeling, such as insurance pricing, indirect or proxy discrimination is an issue of major concern. Namely, there exists the possibility that protected policyholder characteristics are implicitly inferred from non-protected ones by predictive models, and are thus having an undesirable (or illegal) impact on prices. A technical solution to this problem relies on building a best-estimate model using all policyholder characteristics (including protected ones) and then averaging out the protected characteristics for calculating individual prices. However, such approaches require full knowledge of policyholders' protected characteristics, which may in itself be problematic. Here, we address this issue by using a multi-task neural network architecture for claim predictions, which can be trained using only partial information on protected characteristics, and it produces prices that are free from proxy discrimination. We demonstrate the use of the proposed model and we find that its predictive accuracy is comparable to a conventional feedforward neural network (on full information). However, this multi-task network has clearly superior performance in the case of partially missing policyholder information.

GlyNet: a multi-task neural network for predicting protein-glycan interactions.

Advances in diagnostics, therapeutics, vaccines, transfusion, and organ transplantation build on a fundamental understanding of glycan–protein interactions. To aid this, we developed GlyNet, a model that accurately predicts interactions (relative binding strengths) between mammalian glycans and 352 glycan-binding proteins, many at multiple concentrations. For each glycan input, our model produces 1257 outputs, each representing the relative interaction strength between the input glycan and a particular protein sample. GlyNet learns these continuous values using relative fluorescence units (RFUs) measured on 599 glycans in the Consortium for Functional Glycomics glycan arrays and extrapolates these to RFUs from additional, untested glycans. GlyNet's output of continuous values provides more detailed results than the standard binary classification models. After incorporating a simple threshold to transform such continuous outputs the resulting GlyNet classifier outperforms those standard classifiers. GlyNet is the first multi-output regression model for predicting protein–glycan interactions and serves as an important benchmark, facilitating development of quantitative computational glycobiology.

Multi-Task Neural Network for Position Estimation in Large-Scale Indoor Environments

Pedestrian localization within large-scale multi-building/multi-floor indoor environments remains a challenging task. Fingerprinting-based approaches are particularly suited for such large-scale deployments due to their low requirements of hardware installments. Recently, the fingerprinting problem has been addressed by deep learning. Existing models are mostly task specific by providing floor classification or position estimation within a small area. A strategy to support localization within large-scale environments is to sequentially apply hierarchical models. This has several drawbacks including missing scalability and increased deployment complexity on smartphones. We propose a unifying approach based on training a single neural network that classifies the building/floor and predicts the position in a single forward-pass of the network. Our model classifies a grid cell and performs within grid cell regression, which solves the performance degradation of applying regression within large areas. To reduce the error in case of misclassified grid cells, we propose a novel technique called multi cell encoding learning (multi-CEL), where a model simultaneously learns several redundant position representations within an overlapping grid cell encoding. For three public WLAN fingerprinting datasets, we demonstrate that multi-CEL surpasses existing state-of-the-art multi-task learning neural networks and even outperforms regression neural networks explicitly trained for 2D-positioning by up to 17%.

Inversion of equations of state by combining multi-task neural networks and Newton's method

In this work, we address the problem of numerical multidimensional inversion of equations of state for fluid dynamic problems. Particularly, we study the cases where the thermodynamic region of interest contains phase transitions with possible discontinuities of derivatives. In these cases, straight applications of common iterative solvers such as Newton's method with fixed initial guess may lead to divergences. To address this issue, we present a novel method for inverting non-linear equations. The method combines deep multi-task learning techniques with an iterative solver such as Newton's method and allows to handle non-linear equations of complex structure. In order to test the efficiency of the new approach, we apply the method to the problem of inverting equations of state for the pure fluids water, heptane and ethanol. We observe that the proposed method allows to significantly speed-up inversion and avoid divergences, especially in the vicinity of phase boundaries.

Perceptual Image Hashing Based on Multitask Neural Network

With the advent of the era of multimedia and in-depth development, the whole human society has been produced and spread a huge amount of image data, but at the same time, in view of the digital image and tamper with the attack of piracy phenomenon also more and more serious, malicious attacks will produce serious social, military, and political influence, therefore, to protect the authenticity of the original image content, which is also more and more important. In order to further improve the performance of image hashing and enhance the protection of image data, we proposed an end-to-end dual-branch multitask neural network based on VGG-19 to produce a perceptual hash sequence and used prepart of network of pretrained VGG-19 model to extract image features, and then, the image features are transformed into a hash sequence through a convolutional and fully connected network. At the same time, in order to enhance the function of the network and improve the adaptability of the proposed network to using scenarios, the rest part of the network layer of the VGG-19 model was used as another branch for image classification, so as to realize the multitask characteristics of the network. Through the experiment of the testing set, the network can not only resist many kinds of attack operations (content retention operations), but also realize accurate classification about the image, and has a satisfactory tampering detection ability.

A contextual multi-task neural approach to medication and adverse events identification from clinical text

Effective wide-scale pharmacovigilance calls for accurate named entity recognition (NER) of medication entities such as drugs, dosages, reasons, and adverse drug events (ADE) from clinical text. The scarcity of adverse event annotations and underlying semantic ambiguities make accurate scope identification challenging. The current research explores integrating contextualized language models and multi-task learning from diverse clinical NER datasets to mitigate this challenge. We propose a novel multi-task adaptation method to refine the embeddings generated by the Bidirectional Encoder Representations from Transformers (BERT) language model to improve inter-task knowledge sharing. We integrated the adapted BERT model into a unique hierarchical multi-task neural network comprised of the medication and auxiliary clinical NER tasks. We validated the model using two different versions of BERT on diverse well-studied clinical tasks: Medication and ADE (n2c2 2018/n2c2 2009), Clinical Concepts (n2c2 2010/n2c2 2012), Disorders (ShAReCLEF 2013). Overall medication extraction performance enhanced by up to +1.19 F1 (n2c2 2018) while generalization enhanced by +5.38 F1 (n2c2 2009) as compared to standalone BERT baselines. ADE recognition enhanced significantly (McNemar’s test), out-performing prior baselines. Similar benefits were observed on the auxiliary clinical and disorder tasks. We demonstrate that combining multi-dataset BERT adaptation and multi-task learning out-performs prior medication extraction methods without requiring additional features, newer training data, or ensembling. Taken together, the study contributes an initial case study towards integrating diverse clinical datasets in an end-to-end NER model for clinical decision support.

Parsimonious Optimization of Multitask Neural Network Hyperparameters.

Neural networks are rapidly gaining popularity in chemical modeling and Quantitative Structure–Activity Relationship (QSAR) thanks to their ability to handle multitask problems. However, outcomes of neural networks depend on the tuning of several hyperparameters, whose small variations can often strongly affect their performance. Hence, optimization is a fundamental step in training neural networks although, in many cases, it can be very expensive from a computational point of view. In this study, we compared four of the most widely used approaches for tuning hyperparameters, namely, grid search, random search, tree-structured Parzen estimator, and genetic algorithms on three multitask QSAR datasets. We mainly focused on parsimonious optimization and thus not only on the performance of neural networks, but also the computational time that was taken into account. Furthermore, since the optimization approaches do not directly provide information about the influence of hyperparameters, we applied experimental design strategies to determine their effects on the neural network performance. We found that genetic algorithms, tree-structured Parzen estimator, and random search require on average 0.08% of the hours required by grid search; in addition, tree-structured Parzen estimator and genetic algorithms provide better results than random search.

Development and validation of a prediction model for actionable aspects of frailty in the text of clinicians' encounter notes.

Frailty is a prevalent risk factor for adverse outcomes among patients with chronic lung disease. However, identifying frail patients who may benefit from interventions is challenging using standard data sources. We therefore sought to identify phrases in clinical notes in the electronic health record (EHR) that describe actionable frailty syndromes. We used an active learning strategy to select notes from the EHR and annotated each sentence for 4 actionable aspects of frailty: respiratory impairment, musculoskeletal problems, fall risk, and nutritional deficiencies. We compared the performance of regression, tree-based, and neural network models to predict the labels for each sentence. We evaluated performance with the scaled Brier score (SBS), where 1 is perfect and 0 is uninformative, and the positive predictive value (PPV). We manually annotated 155952 sentences from 326 patients. Elastic net regression had the best performance across all 4 frailty aspects (SBS 0.52, 95% confidence interval [CI] 0.49-0.54) followed by random forests (SBS 0.49, 95% CI 0.47-0.51), and multi-task neural networks (SBS 0.39, 95% CI 0.37-0.42). For the elastic net model, the PPV for identifying the presence of respiratory impairment was 54.8% (95% CI 53.3%-56.6%) at a sensitivity of 80%. Classification models using EHR notes can effectively identify actionable aspects of frailty among patients living with chronic lung disease. Regression performed better than random forest and neural network models. NLP-based models offer promising support to population health management programs that seek to identify and refer community-dwelling patients with frailty for evidence-based interventions.

Buzz Tweet Classification Based on Text and Image Features of Tweets Using Multi-Task Learning

This study investigates social media trends and proposes a buzz tweet classification method to explore the factors causing the buzz phenomenon on Twitter. It is difficult to identify the causes of the buzz phenomenon based solely on texts posted on Twitter. It is expected that by limiting the tweets to those with attached images and using the characteristics of the images and the relationships between the text and images, a more detailed analysis than that of with text-only tweets can be conducted. Therefore, an analysis method was devised based on a multi-task neural network that uses both the features extracted from the image and text as input and the buzz class (buzz/non-buzz) and the number of “likes (favorites)” and “retweets (RTs)” as output. The predictions made using a single feature of the text and image were compared with the predictions using a combination of multiple features. The differences between buzz and non-buzz features were analyzed based on the cosine similarity between the text and the image. The buzz class was correctly identified with a correctness rate of approximately 80% for all combinations of image and text features, with the combination of BERT and VGG16 providing the highest correctness rate.

Multitask neural networks for predicting bladder pressure with time series data

Multitask learning (MTL) can improve accuracy over vanilla neural networks in modeling population level time series data. This can be accomplished by assigning the prediction for each individual in the population as a separate task, thereby leveraging the heterogeneity of population level data. Here, we investigate a novel approach by training recurrent neural networks (RNNs) in a multitask setting. We apply this new methodology to experimental data for predicting bladder pressure, and then bladder contractions, from an external urethral sphincter electromyograph (EUS EMG) signal. We found that the multitask models make more accurate individual level predictions than their single tasking counterparts. We observed that, for bladder pressure prediction, either incorporating multitask learning or RNN structure generalized best to out of sample test data and multitasking RNNs had high out of sample correlation coefficients. These results suggest that MTL models could be used to leverage heterogeneous population time series data for making individualized predictions. From these bladder pressure predictions, we predicted the onset of bladder contractions. Our results indicate that the MTL RNN model was superior in both intra- and inter-individual bladder contraction predictions as measured by sensitivity (85.7%), specificity (98.7%) and precision (73.5%).

A Deep Learning and GIS Approach for the Optimal Positioning of Wave Energy Converters

Renewable Energy Sources provide a viable solution to the problem of ever-increasing climate change. For this reason, several countries focus on electricity production using alternative sources. In this paper, the optimal positioning of the installation of wave energy converters is examined taking into account geospatial and technical limitations. Geospatial constraints depend on Land Use classes and seagrass of the coastal areas, while technical limitations include meteorological conditions and the morphology of the seabed. Suitable installation areas are selected after the exclusion of points that do not meet the aforementioned restrictions. We implemented a Deep Neural Network that operates based on heterogeneous data fusion, in this case satellite images and time series of meteorological data. This fact implies the definition of a two-branches architecture. The branch that is trained with image data provides for the localization of dynamic geospatial classes in the potential installation area, whereas the second one is responsible for the classification of the region according to the potential wave energy using wave height and period time series. In making the final decision on the suitability of the potential area, a large number of static land use data play an important role. These data are combined with neural network predictions for the optimizing positioning of the Wave Energy Converters. For the sake of completeness and flexibility, a Multi-Task Neural Network is developed. This model, in addition to predicting the suitability of an area depending on seagrass patterns and wave energy, also predicts land use classes through Multi-Label classification process. The proposed methodology is applied in the marine area of the city of Sines, Portugal. The first neural network achieves 98.7% Binary Classification accuracy, while the Multi-Task Neural Network 97.5% in the same metric and 93.5% in the F1 score of the Multi-Label classification output.

Simulating Player Behavior for Data-Driven Interactive Narrative Personalization

Data-driven approaches to interactive narrative personalization show significant promise for applications in entertainment, training, and education. A common feature of data-driven interactive narrative planning methods is that an enormous amount of training data is required, which is rarely available and expensive to collect from observations of human players. An alternative approach to obtaining data is to generate synthetic data from simulated players. In this paper, we present a long short-term memory (LSTM) neural network framework for simulating players to train data-driven interactive narrative planners. By leveraging a small amount of previously collected human player interaction data, we devise a generative player simulation model. A multi-task neural network architecture is proposed to estimate player actions and experiential outcomes from a single model. Empirical results demonstrate that the bipartite LSTM network produces the better-performing player action prediction models than several baseline techniques, and the multi-task LSTM derives comparable player outcome prediction models within a shorter training time. We also find that synthetic data from the player simulation model contributes to training more effective interactive narrative planners than raw human player data alone.

A fast instance segmentation with one-stage multi-task deep neural network for autonomous driving

An accurate real-time instance segmentation, which can perform both object detection and semantic segmentation at the same time with a multi-task neural network, is important for autonomous driving. This paper proposes a fast one-stage multi-task neural network for instance segmentation, which can meet the requirements of real-time processing with sufficient accuracy and that is more desirable for self-driving applications. With a one-stage strategy, it can perform object detection and segmentation concurrently. This paper conducts the related experiments with two public datasets. The cross-validation was carried out with set of variables to determine the optimal combination of each model and compared with the mainstream instance segmentation algorithms. According to our experiment, the proposed algorithm has five times the performance compared to the previous algorithms, which can meet the real-time requirement for autonomous driving applications.