Multi-task Approach Research Articles

Mitigating Bias in Large Language Models: A Multi-Task Training Approach Using BERT

Abstract. Large language models (LLMs), such as ChatGPT, have become essential tools due to their advanced natural language processing capabilities. However, these models, trained on extensive internet text, can inadvertently learn and propagate unwanted biases, impacting their outputs. This study addresses this issue by analyzing and mitigating such biases through a multi-task and multi-stage training approach. Utilizing the Winograd Bias (Winobias) dataset, the research fine-tunes the Bidirectional Encoder Representations from Transformers (BERT) model to reduce biased outputs. The approach includes an initial mask task to establish a general understanding and a subsequent cloze task to specifically target and mitigate biases. Results demonstrate a significant reduction in bias, with the original model showing approximately 90% certainty in biased outputs, whereas the de-biased model reduced this certainty to 55%. This study effectively showcases a method for bias reduction by modifying only a few parameters, emphasizing a practical approach to enhancing fairness and balance in LLMs used across various applications.

Prospective prediction of childhood body mass index trajectories using multi-task Gaussian processes.

Body mass index (BMI) trajectories have been used to assess the growth of children with respect to their peers, and to anticipate future obesity and disease risk. While retrospective BMI trajectories have been actively studied, models to prospectively predict continuous BMI trajectories have not been investigated. Using longitudinal BMI measurements between birth and age 10 y from a mother-offspring cohort, we leveraged a multi-task Gaussian process approach to develop and evaluate a unified framework for modeling, clustering, and prospective prediction of BMI trajectories. We compared its sensitivity to missing values in the longitudinal follow-up of children, compared its prediction performance to cubic B-spline and multilevel Jenss-Bayley models, and used prospectively predicted BMI trajectories to assess the probability of future BMIs crossing the clinical cutoffs for obesity. MagmaClust identified 5 distinct patterns of BMI trajectories between 0 to 10 y. The method outperformed both cubic B-spline and multilevel Jenss-Bayley models in the accuracy of retrospective BMI trajectories while being more robust to missing data (up to 90%). It was also better at prospectively forecasting BMI trajectories of children for periods ranging from 2 to 8 years into the future, using historic BMI data. Given BMI data between birth and age 2 years, prediction of overweight/obesity status at age 10 years, as computed from MagmaClust's predictions exhibited high specificity (0.94), negative predictive value (0.89), and accuracy (0.86). The accuracy, sensitivity, and positive predictive value of predictions increased as BMI data from additional time points were utilized for prediction. MagmaClust provides a unified, probabilistic, non-parametric framework to model, cluster, and prospectively predict childhood BMI trajectories and overweight/obesity risk. The proposed method offers a convenient tool for clinicians to monitor BMI growth in children, allowing them to prospectively identify children with high predicted overweight/obesity risk and implement timely interventions.

A multi-task genetic programming approach for online multi-objective container placement in heterogeneous cluster

A multi-task genetic programming approach for online multi-objective container placement in heterogeneous cluster

MolE: a foundation model for molecular graphs using disentangled attention.

Models that accurately predict properties based on chemical structure are valuable tools in the chemical sciences. However, for many properties, public and private training sets are typically small, making it difficult for models to generalize well outside of the training data. Recently, this lack of generalization has been mitigated by using self-supervised pretraining on large unlabeled datasets, followed by finetuning on smaller, labeled datasets. Inspired by these advances, we report MolE, a Transformer architecture adapted for molecular graphs together with a two-step pretraining strategy. The first step of pretraining is a self-supervised approach focused on learning chemical structures trained on ~842 million molecular graphs, and the second step is a massive multi-task approach to learn biological information. We show that finetuning models that were pretrained in this way perform better than the best published results on 10 of the 22 ADMET (absorption, distribution, metabolism, excretion and toxicity) tasks included in the Therapeutic Data Commons leaderboard (c. September 2023).

Multi-task deep learning for quantifying methane emissions from 2-D plume imagery with Low Signal-to-Noise Ratio

ABSTRACT Methane is the second most significant greenhouse gas after carbon dioxide. As global warming intensifies, the quantification of methane point sources is becoming increasingly crucial. However, retrieving high-quality methane signals from remote sensing data remains challenging due to various factors, including surface reflectance, atmospheric interference, sensor noise, wind speed, and sensor sensitivity. In real-world scenarios, methane remote sensing quantification often encounters unfavourable conditions that lead to low signal-to-noise ratio (SNR) signals, resulting in reduced quantification accuracy. To address these challenges, we introduce a novel multi-task learning-based approach. Specifically, we incorporate a denoising auxiliary task into the quantification network by introducing an additional denoising branch that recovers clean plume column concentration maps. The network is trained with supervision using both denoising and quantification losses, which enables it to acquire robust feature representations to noise and benefits the quantification of low SNR images. During the inference phase, the denoising branch is removed, resulting in an efficient and robust single quantification network with reduced inferring time, supporting onboard computation. We construct a low SNR database based on the AVIRIS-NG sensor and evaluate the generalization ability of our method. DQNet achieves the highest RMSE, MAPE, and R of 16.478 kg/h, 15.296%, and 95.167% respectively on the synthetic dataset. Across the entire range of SNR, DQNet exhibits a greater relative advantage as SNR decreases. However, our approach is not limited to AVIRIS-NG and can be easily extended to various multispectral and hyperspectral satellites.

Enfoque multitareas implementado en una minicomputadora para el control y monitoreo de un motor a pasos industrial

This paper presents the implementation of the multitasking approach on a Raspberry Pi single board minicomputer, with experimental results, of the control and monitoring of an industrial type Nema 34 stepper motor through a user interface that allows the configuration of the control operation and data visualization to interact with the physical devices. One of the tasks consists of generating a pulse width modulation that is sent to the DQ860HA driver to rotate the stepper motor considering 400 pulses per revolution, while the second task simultaneously counts the number of pulses produced by an encoder with a resolution of 100 pulses per revolution, which is coupled to the motor shaft. The use of low-cost computer equipment and open-source software makes the presented development can be considered as a frugal technology application for the control and monitoring of an industrial system, with experimental results that validate the correct operation of two tasks that must be executed simultaneously.

Automated segment-level coronary artery calcium scoring on non-contrast CT: a multi-task deep-learning approach

ObjectivesTo develop and evaluate a multi-task deep-learning (DL) model for automated segment-level coronary artery calcium (CAC) scoring on non-contrast computed tomography (CT) for precise localization and quantification of calcifications in the coronary artery tree.MethodsThis study included 1514 patients (mean age, 60.0 ± 10.2 years; 56.0% female) with stable chest pain from 26 centers participating in the multicenter DISCHARGE trial (NCT02400229). The patients were randomly assigned to a training/validation set (1059) and a test set (455). We developed a multi-task neural network for performing the segmentation of calcifications on the segment level as the main task and the segmentation of coronary artery segment regions with weak annotations as an auxiliary task. Model performance was evaluated using (micro-average) sensitivity, specificity, F1-score, and weighted Cohen’s κ for segment-level agreement based on the Agatston score and performing interobserver variability analysis.ResultsIn the test set of 455 patients with 1797 calcifications, the model assigned 73.2% (1316/1797) to the correct coronary artery segment. The model achieved a micro-average sensitivity of 0.732 (95% CI: 0.710–0.754), a micro-average specificity of 0.978 (95% CI: 0.976–0.980), and a micro-average F1-score of 0.717 (95% CI: 0.695–0.739). The segment-level agreement was good with a weighted Cohen’s κ of 0.808 (95% CI: 0.790–0.824), which was only slightly lower than the agreement between the first and second observer (0.809 (95% CI: 0.798–0.845)).ConclusionAutomated segment-level CAC scoring using a multi-task neural network approach showed good agreement on the segment level, indicating that DL has the potential for automated coronary artery calcification classification.Critical relevance statementMulti-task deep learning can perform automated coronary calcium scoring on the segment level with good agreement and may contribute to the development of new and improved calcium scoring methods.Key PointsSegment-level coronary artery calcium scoring is a tedious and error-prone task.The proposed multi-task model achieved good agreement with a human observer on the segment level.Deep learning can contribute to the automation of segment-level coronary artery calcium scoring.Graphical

Efficient self-attention based joint optimization for lithology and petrophysical parameter estimation in the Athabasca Oil Sands

Accurately identifying lithology and petrophysical parameters, such as porosity and water saturation, are essential in reservoir characterization. Manual interpretation of well-log data, the conventional approach, is not only labor-intensive but also susceptible to human errors. To address these challenges of lithology identification and petrophysical parameter estimation in the Athabasca Oil Sands area, this study introduces an AutoRegressive Vision Transformer (ARViT) model for lithology and petrophysical parameter prediction. The effectiveness of ARViT lies in its self-attention mechanism and its ability to handle data sequentially, allowing the model to capture important spatial dependencies within the well-log data. This mechanism enables the model to identify subtle spatial and temporal relationships among various geophysical measurements. The model is also interpretable and can serve as an assistive tool for geoscientists, enabling faster interpretation while reducing human bias. The interpretable nature of the model should assist geoscientists in conducting faster quality checks of the predictions, ensuring that errors are not propagated to subsequent stages. This study adopts a multitask learning approach, jointly optimizing the model's performance across multiple tasks simultaneously. To evaluate the effectiveness of the ARViT model, we conducted series of experiments and comparisions, testing it against traditional artificial neural networks (ANN), Long Short-Term Memory (LSTM), and Vision Transformer (ViT) models. To showcase the versatility of ARViT, we apply Low-Rank Adaptation (LoRA) to a different smaller dataset, showing its potential to adapt to different geological contexts. LoRA not only helps in model adaptability but also helps to reduce the number of trainable parameters. Our findings demonstrate that ARViT outperforms ANN, LSTM, and ViT in estimating lithological and petrophysical parameters. While lithology prediction has been a well-explored field, ARViT's unique blend of features, including its self-attention mechanism, autoregression, and multitask approach along with efficient fine tuning using LoRA, sets it apart as a valuable tool for the complex task of lithology prediction and petrophysical parameter estimation.

Multi-microphone simultaneous speakers detection and localization of multi-sources for separation and noise reduction

This paper addresses the challenge of online blind speaker separation in a multi-microphone setting. The linearly constrained minimum variance (LCMV) beamformer is selected as the backbone of the separation algorithm due to its distortionless response and capacity to create a null towards interfering sources. A specific instance of the LCMV beamformer that considers acoustic propagation is implemented. In this variant, the relative transfer functions (RTFs) associated with each speaker of interest are utilized as the steering vectors of the beamformer. A control mechanism is devised to ensure robust estimation of the beamformer’s building blocks, comprising speaker activity detectors and direction of arrival (DOA) estimation branches. This control mechanism is implemented as a multi-task deep neural network (DNN). The primary task classifies each time frame based on speaker activity: no active speaker, single active speaker, or multiple active speakers. The secondary task is DOA estimation. It is implemented as a classification task, executed only for frames classified as single-speaker frames by the primary branch. The direction of the active speaker is classified into one of the multiple ranges of angles. These frames are also leveraged to estimate the RTFs using subspace estimation methods. A library of RTFs associated with these DOA ranges is then constructed, facilitating rapid acquisition of new speakers and efficient tracking of existing speakers. The proposed scheme is evaluated in both simulated and real-life recordings, encompassing static and dynamic scenarios. The benefits of the multi-task approach are showcased, and significant improvements are evident, even when the control mechanism is trained with simulated data and tested with real-life data. A comparison between the proposed scheme and the independent low-rank matrix analysis (ILRMA) algorithm reveals significant improvements in static scenarios. Furthermore, the tracking capabilities of the proposed scheme are highlighted in dynamic scenarios.

A comparative study of evolutionary algorithms and particle swarm optimization approaches for constrained multi-objective optimization problems

Many real-world optimization problems contain multiple conflicting objectives as well as additional problem constraints. These problems are referred to as constrained multi-objective optimization problems (CMOPs). Many meta-heuristics for solving CMOPs, called constrained multi-objective meta-heuristics (CMOMHs) have been introduced in the literature, including those using particle swarm optimization (PSO)(Kennedy and Eberhart, 1995), genetic algorithms (GAs)(Man et al., 1996), and differential evolution (DE)(Storn and Price, 1997). CMOMHs can be grouped into four different classes: classic CMOMHs, co-evolutionary approaches, multi-stage approaches, and multi-tasking approaches. An extensive comparative study of twenty different CMOMHs on a wide variety of test problems, including real-world CMOPs in the fields of science and engineering, is conducted. A multi-swarm PSO approach called constrained multi-guide particle swarm optimization (ConMGPSO) is introduced and compared to the best-performing previous approaches according to the comparative study. The performance of each algorithm was found to be problem dependent, however the best overall approaches were ConMGPSO, paired-offspring constrained evolutionary algorithm (POCEA)(He et al., 2021), adaptive non-dominated sorting genetic algorithm III (A-NSGA-III)(Jain and Deb, 2014), and constrained multi-objective framework using Q-learning and evolutionary multi-tasking (CMOQLMT)(Ming and Gong, 2023). ConMGPSO and POCEA had the best performance on the CF benchmark set, which contains examples of bi-objective and tri-objective CMOPs with disconnected CPOFs. The CMOQLMT approach had the best performance on the DAS-CMOP benchmark set, which contain additional difficulty in terms of feasibility-, convergence-, and diversity-hardness. For the selected real-world CMOPs, A-NSGA-III had the best performance overall. ConMGPSO was shown to have the best performance on the process, design, and synthesis problems, and had competitive performance for the power system optimization problems.

Learning to explain is a good biomedical few-shot learner.

Significant progress has been achieved in biomedical text mining using deep learning methods, which rely heavily on large amounts of high-quality data annotated by human experts. However, the reality is that obtaining high-quality annotated data is extremely challenging due to data scarcity (e.g. rare or new diseases), data privacy and security concerns, and the high cost of data annotation. Additionally, nearly all researches focus on predicting labels without providing corresponding explanations. Therefore, in this paper, we investigate a more realistic scenario, biomedical few-shot learning, and explore the impact of interpretability on biomedical few-shot learning. We present LetEx-Learning to explain-a novel multi-task generative approach that leverages reasoning explanations from large language models (LLMs) to enhance the inductive reasoning ability of few-shot learning. Our approach includes (1) collecting high-quality explanations by devising a suite of complete workflow based on LLMs through CoT prompting and self-training strategies, (2) converting various biomedical NLP tasks into a text-to-text generation task in a unified manner, where collected explanations serve as additional supervision between text-label pairs by multi-task training. Experiments are conducted on three few-shot settings across six biomedical benchmark datasets. The results show that learning to explain improves the performances of diverse biomedical NLP tasks in low-resource scenario, outperforming strong baseline models significantly by up to 6.41%. Notably, the proposed method makes the 220M LetEx perform superior reasoning explanation ability against LLMs. Our source code and data are available at https://github.com/cpmss521/LetEx.

Multi-View Integrative Approach For Imputing Short-Chain Fatty Acids and Identifying Key factors predicting Blood SCFA.

Short-chain fatty acids (SCFAs) are the main metabolites produced by bacterial fermentation of dietary fiber within gastrointestinal tract. SCFAs produced by gut microbiotas (GMs) are absorbed by host, reach bloodstream, and are distributed to different organs, thus influencing host physiology. However, due to the limited budget or the poor sensitivity of instruments, most studies on GMs have incomplete blood SCFA data, limiting our understanding of the metabolic processes within the host. To address this gap, we developed an innovative multi-task multi-view integrative approach (M2AE, Multi-task Multi-View Attentive Encoders), to impute blood SCFA levels using gut metagenomic sequencing (MGS) data, while taking into account the intricate interplay among the gut microbiome, dietary features, and host characteristics, as well as the nuanced nature of SCFA dynamics within the body. Here, each view represents a distinct type of data input (i.e., gut microbiome compositions, dietary features, or host characteristics). Our method jointly explores both view-specific representations and cross-view correlations for effective predictions of SCFAs. We applied M2AE to two in-house datasets, which both include MGS and blood SCFAs profiles, host characteristics, and dietary features from 964 subjects and 171 subjects, respectively. Results from both of two datasets demonstrated that M2AE outperforms traditional regression-based and neural-network based approaches in imputing blood SCFAs. Furthermore, a series of gut bacterial species (e.g., Bacteroides thetaiotaomicron and Clostridium asparagiforme), host characteristics (e.g., race, gender), as well as dietary features (e.g., intake of fruits, pickles) were shown to contribute greatly to imputation of blood SCFAs. These findings demonstrated that GMs, dietary features and host characteristics might contribute to the complex biological processes involved in blood SCFA productions. These might pave the way for a deeper and more nuanced comprehension of how these factors impact human health.

An improved multi-task approach for SHM missing data reconstruction using attentive neural process and meta-learning

An improved multi-task approach for SHM missing data reconstruction using attentive neural process and meta-learning

Multi-Task Intelligent Monitoring of Construction Safety Based on Computer Vision

Effective safety management is vital for ensuring construction safety. Traditional safety inspections in construction heavily rely on manual labor, which is both time-consuming and labor-intensive. Extensive research has been conducted integrating computer-vision technologies to facilitate intelligent surveillance and improve safety measures. However, existing research predominantly focuses on singular tasks, while construction environments necessitate comprehensive analysis. This study introduces a multi-task computer vision technology approach for the enhanced monitoring of construction safety. The process begins with the collection and processing of multi-source video surveillance data. Subsequently, YOLOv8, a deep learning-based computer vision model, is adapted to meet specific task requirements by modifying the head component of the framework. This adaptation enables efficient detection and segmentation of construction elements, as well as the estimation of person and machine poses. Moreover, a tracking algorithm integrates these capabilities to continuously monitor detected elements, thereby facilitating the proactive identification of unsafe practices on construction sites. This paper also presents a novel Integrated Excavator Pose (IEP) dataset designed to address the common challenges associated with different single datasets, thereby ensuring accurate detection and robust application in practical scenarios.

The Use of Immersive Virtual Reality Training for Developing Nontechnical Skills Among Nursing Students: Multimethods Study.

Immersive virtual reality (IVR) is a niche technology rising in popularity in nursing education. Although there is an abundance of evidence to demonstrate the effect of virtual reality (VR) on desired learning outcomes, this evidence is limited to technical or procedural skills or managing a single patient with clinical problems. Nontechnical skills (NTS), such as communication, decision-making, teamwork, situation awareness, and managerial skills, have not been explored using IVR technology. This study aimed to (1) investigate the potential efficacy of the IVR system virtual reality hospital (VR-Hospital, or VR-Hosp), a single-user game we developed, on nursing students' NTS, sense of presence in the virtual clinical environment, and satisfaction and self-confidence in learning; (2) identify variables that predict NTS; and (3) explore students' experience in using VR-Hosp. A multimethods design with a quantitative and qualitative approach was adopted. Participants were provided with VR-Hosp with 3 scenarios in training. VR-Hosp adopted a multibed, multipatient, multitask approach and was embedded with various clinical situations. Learning outcomes were measured after the training, followed by group interviews. In total, 202 students joined the study. Results revealed high levels of satisfaction and self-confidence in learning. Significant achievement in NTS was perceived by the students. The levels of satisfaction and self-confidence in learning and the involvement and sensory fidelity domains in the sense of presence were positive predictors of NTS. The promising results offer a basis for designing IVR activities for nursing education. Further investigations are imperative to determine the impact of IVR technology on learning outcomes in clinical practice.

Multi-domain awareness for compressed deepfake videos detection over social networks guided by common mechanisms between artifacts

The viral spread of massive deepfake videos over social networks has caused serious security problems. Despite the remarkable advancements achieved by existing deepfake detection algorithms, deepfake videos over social networks are inevitably influenced by compression factors. This causes deepfake detection performance to be limited by the following challenging issues: (a) interfering with compression artifacts, (b) loss of feature information, and (c) aliasing of feature distributions. In this paper, we analyze the common mechanism between compression artifacts and deepfake artifacts, revealing the structural similarity between them and providing a reliable theoretical basis for enhancing the robustness of deepfake detection models against compression. Firstly, based on the common mechanism between artifacts, we design a frequency domain adaptive notch filter to eliminate the interference of compression artifacts on specific frequency bands. Secondly, to reduce the sensitivity of deepfake detection models to unknown noise, we propose a spatial residual denoising strategy. Thirdly, to exploit the intrinsic correlation between feature vectors in the frequency domain branch and the spatial domain branch, we enhance deepfake features using an attention-based feature fusion method. Finally, we adopt a multi-task decision approach to enhance the discriminative power of the latent space representation of deepfakes, achieving deepfake detection with robustness against compression. Extensive experiments show that compared with the baseline methods, the detection performance of the proposed algorithm on compressed deepfake videos has been significantly improved. In particular, our model is resistant to various types of noise disturbances and can be easily combined with baseline detection models to improve their robustness.

From sight to insight: A multi-task approach with the visual language decoding model

Visual neural decoding aims to unlock the mysteries of how the human brain interprets the visual world through predicting perceived visual information from visual neural activity. While early studies made some progress in decoding visual activity for singular type of information, they failed to concurrently reveal the multi-level interweaving linguistic information in the brain. Here, we developed a novel Visual Language Decoding Model (VLDM) that can simultaneously decode the main categories, semantic labels, and textual descriptions of visual stimuli from visual activities. The large-scale NSD dataset was utilized to ensure the efficiency of the decoding model in joint training and evaluation across multiple tasks. For category decoding, we achieved the effective classification of 12 categories with an accuracy of nearly 70 %, significantly surpassing the chance level. For label decoding, we attained the precise prediction of 80 specific semantic labels with a 16-fold improvement over the chance level. For text decoding, the scores of the decoded text surpassed the corresponding baseline levels by remarkable margins on six evaluation metrics. These results highlight the complexity of how the brain processes visual information and the close connection between visual perception and language cognition. This study contributes significantly to extensive applications in multi-layered brain-computer interfaces, potentially leading to more natural and efficient human-computer interaction experiences.

Machine learning prediction of malaria vaccine efficacy based on antibody profiles.

Immunization through repeated direct venous inoculation of Plasmodium falciparum (Pf) sporozoites (PfSPZ) under chloroquine chemoprophylaxis, using the PfSPZ Chemoprophylaxis Vaccine (PfSPZ-CVac), induces high-level protection against controlled human malaria infection (CHMI). Humoral and cellular immunity contribute to vaccine efficacy but only limited information about the implicated Pf-specific antigens is available. Here, we examined Pf-specific antibody profiles, measured by protein arrays representing the full Pf proteome, of 40 placebo- and PfSPZ-immunized malaria-naïve volunteers from an earlier published PfSPZ-CVac dose-escalation trial. For this purpose, we both utilized and adapted supervised machine learning methods to identify predictive antibody profiles at two different time points: after immunization and before CHMI. We developed an adapted multitask support vector machine (SVM) approach and compared it to standard methods, i.e. single-task SVM, regularized logistic regression and random forests. Our results show, that the multitask SVM approach improved the classification performance to discriminate the protection status based on the underlying antibody-profiles while combining time- and dose-dependent data in the prediction model. Additionally, we developed the new fEature diStance exPlainabilitY (ESPY) method to quantify the impact of single antigens on the non-linear multitask SVM model and make it more interpretable. In conclusion, our multitask SVM model outperforms the studied standard approaches in regard of classification performance. Moreover, with our new explanation method ESPY, we were able to interpret the impact of Pf-specific antigen antibody responses that predict sterile protective immunity against CHMI after immunization. The identified Pf-specific antigens may contribute to a better understanding of immunity against human malaria and may foster vaccine development.

Automatic question-answer pairs generation using pre-trained large language models in higher education

The process of manually generating question and answer (QA) pairs for assessments is known to be a time-consuming and energy-intensive task for teachers, specifically in higher education. Several studies have proposed various methods utilising pre-trained large language models for the generation of QA pairs. However, it is worth noting that these methods have primarily been evaluated on datasets that are not specifically educational in nature. Furthermore, the evaluation metrics and strategies employed in these studies differ significantly from those typically used in educational contexts. The present discourse fails to present a compelling case regarding the efficacy and practicality of stated methods within the context of higher education. This study aimed to examine multiple QA pairs generation approaches in relation to their performance and the efficacy and constraints within the context of higher education. The various approaches encompassed in this study comprise pipeline, joint, multi-task approach. The performance of these approaches under consideration was assessed on three datasets related to distinct courses. The evaluation integrates three automated methods, teacher assessments, and real-world educational evaluations to provide a comprehensive analysis. The comparison of various approaches was conducted by directly assessing their performance using the average scores of different automatic metrics on three datasets. The results of the teachers and real educational evaluation indicate that the assessments generated were beneficial in enhancing the understanding of concepts and overall performance of students. The implications of the findings from this study hold significant importance in enhancing the efficacy of QA pair generation tools within the context of higher education.

Completion of partial chemical equations

Inferring missing molecules in chemical equations is an important task in chemistry and drug discovery. In fact, the completion of chemical equations with necessary reagents is important for improving existing datasets by detecting missing compounds, making them compatible with deep learning models that require complete information about reactants, products, and reagents in a chemical equation for increased performance. Here, we present a deep learning model to predict missing molecules using a multi-task approach, which can ultimately be viewed as a generalization of the forward reaction prediction and retrosynthesis models, since both can be expressed in terms of incomplete chemical equations. We illustrate that a single trained model, based on the transformer architecture and acting on reaction SMILES strings, can address the prediction of products (forward), precursors (retro) or any other molecule in arbitrary positions such as solvents, catalysts or reagents (completion). Our aim is to assess whether a unified model trained simultaneously on different tasks can effectively leverage diverse knowledge from various prediction tasks within the chemical domain, compared to models trained individually on each application. The multi-task models demonstrate top-1 performance of 72.4%, 16.1%, and 30.5% for the forward, retro, and completion tasks, respectively. For the same model we computed round-trip accuracy of 83.4%. The completion task exhibiting improvements due to the multi-task approach.