Explore the vast range of titles available.

Multi-task learning (MTL), which optimizes multiple related learning tasks at the same time, has been widely used in various applications, including natural language processing, speech recognition, computer vision, multimedia data processing, biomedical imaging, socio-biological data analysis, multi-modality data analysis, etc. MTL sometimes is also referred to as joint learning, and is closely related to other machine learning subfields like multi-class learning, transfer learning, and learning with auxiliary tasks, to name a few. In this paper, we provide a brief review on this topic, discuss the motivation behind this machine learning method, compare various MTL algorithms, review MTL methods for incomplete data, and discuss its application in deep learning. We aim to provide the readers with a simple way to understand MTL without too many complicated equations, and to help the readers to apply MTL in their applications.

Multi-Task Learning (MTL) is a powerful technique that has gained popularity due to its performance improvement over traditional Single-Task Learning (STL). However, MTL is often challenging because there is an exponential number of possible task groupings, which can make it difficult to choose the best one because some groupings might produce performance degradation due to negative interference between tasks. That is why existing solutions are severely suffering from scalability issues, limiting any practical application. In our paper, we propose a new data-driven method that addresses these challenges and provides a scalable and modular solution for classification task grouping based on a re-proposed data-driven features, Data Maps, which capture the training dynamics for each classification task during the MTL training. Through a theoretical comparison with other techniques, we manage to show that our approach has the superior scalability. Our experiments show a better performance and verify the method’s effectiveness, even on an unprecedented number of tasks (up to 100 tasks on CIFAR100). Being the first to work on such number of tasks, our comparisons on the resulting grouping shows similar grouping to the mentioned in the dataset, CIFAR100. Finally, we provide a modular implementation 3 3 https://github.com/ammarSherif/STG-MTL . for easier integration and testing, with examples from multiple datasets and tasks.

When stimuli are learned by repetition, they are remembered better and retained for a longer time. However, current findings are lacking as to whether the medial temporal lobe (MTL) and cortical regions are involved in the learning effect when subjects retrieve associative memory, and whether their activations differentially change over time due to learning experience. To address these issues, we designed an fMRI experiment in which face-scene pairs were learned once (L1) or six times (L6). Subjects learned the pairs at four retention intervals, 30-min, 1-day, 1-week and 1-month, after which they finished an associative recognition task in the scanner. The results showed that compared to learning once, learning six times led to stronger activation in the hippocampus, but weaker activation in the perirhinal cortex (PRC) as well as anterior ventrolateral prefrontal cortex (vLPFC). In addition, the hippocampal activation was positively correlated with that of the parahippocampal place area (PPA) and negatively correlated with that of the vLPFC when the L6 group was compared to the L1 group. The hippocampal activation decreased over time after L1 but remained stable after L6. These results clarified how the hippocampus and cortical regions interacted to support associative memory after different learning experiences.

Medial temporal lobe (MTL) atrophy is a key feature of Alzheimer's disease (AD), however, it also occurs in typical aging. To enhance the clinical utility of this biomarker, we need to better understand the differential effects of age and AD by encompassing the full AD-continuum from cognitively unimpaired (CU) to dementia, including all MTL subregions with up-to-date approaches and using longitudinal designs to assess atrophy more sensitively. Age-related trajectories were estimated using the best-fitted polynomials in 209 CU adults (aged 19–85). Changes related to AD were investigated among amyloid-negative (Aβ−) ( n = 46) and amyloid-positive (Aβ+) ( n = 14) CU, Aβ+ patients with mild cognitive impairment (MCI) ( n = 33) and AD ( n = 31). Nineteen MCI-to-AD converters were also compared with 34 non-converters. Relationships with cognitive functioning were evaluated in 63 Aβ+ MCI and AD patients. All participants were followed up to 47 months. MTL subregions, namely, the anterior and posterior hippocampus (aHPC/pHPC), entorhinal cortex (ERC), Brodmann areas (BA) 35 and 36 [as perirhinal cortex (PRC) substructures], and parahippocampal cortex (PHC), were segmented from a T1-weighted MRI using a new longitudinal pipeline (LASHiS). Statistical analyses were performed using mixed models. Adult lifespan models highlighted both linear (PRC, BA35, BA36, PHC) and nonlinear (HPC, aHPC, pHPC, ERC) trajectories. Group comparisons showed reduced baseline volumes and steeper volume declines over time for most of the MTL subregions in Aβ+ MCI and AD patients compared to Aβ− CU, but no differences between Aβ− and Aβ+ CU or between Aβ+ MCI and AD patients (except in ERC). Over time, MCI-to-AD converters exhibited a greater volume decline than non-converters in HPC, aHPC, and pHPC. Most of the MTL subregions were related to episodic memory performances but not to executive functioning or speed processing. Overall, these results emphasize the benefits of studying MTL subregions to distinguish age-related changes from AD. Interestingly, MTL subregions are unequally vulnerable to aging, and those displaying non-linear age-trajectories, while not damaged in preclinical AD (Aβ+ CU), were particularly affected from the prodromal stage (Aβ+ MCI). This volume decline in hippocampal substructures might also provide information regarding the conversion from MCI to AD-dementia. All together, these findings provide new insights into MTL alterations, which are crucial for AD-biomarkers definition.

The manual assessment of Left Ventricular (LV) structure and its systolic function from echocardiography is fundamental for cardiovascular diagnosis but is often time-consuming and subject to inter-observer variability. While deep learning has advanced the automation of individual echocardiographic tasks, the prevailing approach of developing separate models for functional and structural analysis fails to leverage the intrinsic relationship between these two aspects of cardiac health. We propose a novel, unified Multi-Task Learning (MTL) framework designed to simultaneously perform LV segmentation and anatomical keypoint detection from a single analysis. The model employs a shared EfficientNet encoder that feeds into two parallel, specialized heads, including a U-Net-style decoder for segmentation and a convolutional head for heatmap-based keypoint localization. The framework was trained and validated on three large-scale public datasets: CAMUS, EchoNet-Dynamic, and EchoNet-LVH. Our proposed framework achieved state-of-the-art performance on both tasks. For segmentation, the model reached a Dice Similarity Coefficient (DSC) of up to 0.951 on the CAMUS dataset and 0.931 on EchoNet-Dynamic. For keypoint detection, it achieved a low Mean Absolute Error (MAE) of ~ 1.13 pixels across all structural measurements on the EchoNet-LVH dataset. An ablation study also confirmed that the MTL approach synergistically improved the performance of both tasks compared to single-task models. By unifying segmentation with heatmap-based keypoint detection, this synergistic approach offers an efficient, accurate, and interpretable geometry-based alternative to existing systems that rely on direct regression or complex full-wall segmentation 

•We analyze episodic memory-sensitive neurons in the human medial temporal lobe.•Firing patterns of these neurons reinstated at recollection consistent with previous reports.•The magnitude of reinstatement predicted temporal clustering behavior.•These neurons exhibited phase locking to hippocampal theta oscillations.•Spiking at encoding versus retrieval exhibited consistent phase offset. Episodic memory requires associating items with temporal context, a process for which the medial temporal lobe (MTL) is critical. This study uses recordings from 27 human subjects who were undergoing surgical intervention for intractable epilepsy. These same data were also utilized in Umbach et al. (2020). We identify 103 memory-sensitive neurons in the hippocampus and entorhinal cortex, whose firing rates predicted successful episodic memory encoding as subjects performed a verbal free recall task. These neurons exhibit important properties. First, as predicted from the temporal context model, they demonstrate reinstatement of firing patterns observed during encoding at the time of retrieval. The magnitude of reinstatement predicted the tendency of subjects to cluster retrieved memory items according to input serial position. Also, we found that spiking activity of these neurons was locked to the phase of hippocampal theta oscillations, but that the mean phase of spiking shifted between memory encoding versus retrieval. This unique observation is consistent with predictions of the “Separate Phases at Encoding And Retrieval (SPEAR)” model. Together, the properties we identify for memory-sensitive neurons characterize direct electrophysiological mechanisms for the representation of contextual information in the human MTL.

Wind power prediction for newly built wind farms is usually faced with the problem of no sufficient historical data. To efficiently extract the useful features from related wind farms, a novel transfer learning method based on temporal convolutional network (TCN)‐Bi‐long short‐term memory (LSTM) with dynamic loss weights is proposed. Firstly, a novel multi‐task TCN‐Bi‐LSTM model is designed to extract common features. The separate TCNs, and common Bi‐LSTM layers of the proposed model are designed to extract the temporal features from related wind farms. Secondly, in the pre‐training stage, to optimize the training process of the neural networks, a dynamic loss‐weighting strategy is proposed for multi‐task learning (MTL) to select the most related features, which increase the prediction accuracy by providing a suitable optimization object. Thirdly, the multi‐task TCN‐Bi‐LSTM model is re‐trained based on the samples from the target wind farm. Finally, a dataset of seven wind farms was employed to evaluate the efficiency of the proposed MTL structure and the dynamic loss‐weighting strategy. The result shows that the root mean squared error of the 12‐h short‐term prediction can be decreased by 4.19% compared with the traditional single‐task learning model, which verifies the validity of the proposed multi‐task transfer learning method.

This work presents a frequency-domain and modal-domain model to analyze how the length of a three-phase power cable influences conducted emission (CE) voltages measured through a line impedance stabilization network (LISN). The measurement setup considered consists of an equipment under test (EUT) connected to the LISN via a power cable whose cross-section is defined in this study as quadrilateral, namely, four conductors arranged at the corners of a quadrilateral: typically the three phases and the protective earth or neutral conductor. The cable is modeled as a multiconductor transmission line (MTL). To evaluate the system performance both with and without the cable, the concept of voltage insertion ratio (IR) is introduced, defined as the reciprocal of the typical insertion loss. Closed-form expressions are derived for both common mode (CM) and differential mode (DM) emissions. The objective is twofold: to understand under which conditions the LISN measurements overestimate or underestimate the actual emissions at the EUT terminals, and to provide a predictive tool to assess the impact of electrically long cables on CE measurements. The model is validated through numerical simulations of quadrilateral cable configurations considering both a homogeneous and inhomogeneous cross-section, highlighting the need to account for cable length in system design and EMC test interpretation.

While extensive research on the neurophysiology of spatial memory has been carried out in rodents, memory research in humans had traditionally focused on more abstract, language-based tasks. Recent studies have begun to address this gap using virtual navigation tasks in combination with electrophysiological recordings in humans. These studies suggest that the human medial temporal lobe (MTL) is equipped with a population of place and grid cells similar to that previously observed in the rodent brain. Furthermore, theta oscillations have been linked to spatial navigation and, more specifically, to the encoding and retrieval of spatial information. While some studies suggest a single navigational theta rhythm which is of lower frequency in humans than rodents, other studies advocate for the existence of two functionally distinct delta-theta frequency bands involved in both spatial and episodic memory. Despite the general consensus between rodent and human electrophysiology, behavioral work in humans does not unequivocally support the use of a metric Euclidean map for navigation. Formal models of navigational behavior, which specifically consider the spatial scale of the environment and complementary learning mechanisms, may help to better understand different navigational strategies and their neurophysiological mechanisms. Finally, the functional overlap of spatial and declarative memory in the MTL calls for a unified theory of MTL function. Such a theory will critically rely upon linking task-related phenomena at multiple temporal and spatial scales. Understanding how single cell responses relate to ongoing theta oscillations during both the encoding and retrieval of spatial and non-spatial associations appears to be key toward developing a more mechanistic understanding of memory processes in the MTL.

In vivo doubled haploid (DH) production based on crossing heterozygous germplasm with mtl haploid inducer lines promises to transform modern rice (Oryza sativa) breeding. However, this technology is limited, as haploid inducers and pollen acceptors have asynchronous heading dates. To address this obstacle, we developed a panel of multiple-heading-date mtl haploid inducer lines that produce pollen for more than 35 days. We edited the MTL gene in a hybrid rice with the CRISPR-Cas9 system. We then selected transgene-free homozygous mutants in the T1 generation and reproduced to T4 generation by single-seed descent method. We obtained 547 mtl haploid inducers with diverse heading dates (from 73 to 110 days) and selected 16 lines comprising a core population with continuous flowering. The seed-setting rate and haploid induction rate (HIR) of the core panel were 4.0–12.7% and 2.8–12.0%, respectively. Thus, our strategy of using multiple-heading-date mtl haploid inducers could accelerate the use of in vivo DH technology in rice breeding.