Mathematical properties of neuronal TD-rules and differential Hebbian learning: a comparison

A confusingly wide variety of temporally asymmetric learning rules exists related to reinforcement learning and/or to spike-timing dependent plasticity, many of which look exceedingly similar, while displaying strongly different behavior. These rules often find their use in control tasks, for example in robotics and for this rigorous convergence and numerical stability is required. The goal of this article is to review these rules and compare them to provide a better overview over their different properties. Two main classes will be discussed: temporal difference (TD) rules and correlation based (differential hebbian) rules and some transition cases. In general we will focus on neuronal implementations with changeable synaptic weights and a time-continuous representation of activity. In a machine learning (non-neuronal) context, for TD-learning a solid mathematical theory has existed since several years. This can partly be transfered to a neuronal framework, too. On the other hand, only now a more complete theory has also emerged for differential Hebb rules. In general rules differ by their convergence conditions and their numerical stability, which can lead to very undesirable behavior, when wanting to apply them. For TD, convergence can be enforced with a certain output condition assuring that the δ-error drops on average to zero (output control). Correlation based rules, on the other hand, converge when one input drops to zero (input control). Temporally asymmetric learning rules treat situations where incoming stimuli follow each other in time. Thus, it is necessary to remember the first stimulus to be able to relate it to the later occurring second one. To this end different types of so-called eligibility traces are being used by these two different types of rules. This aspect leads again to different properties of TD and differential Hebbian learning as discussed here. Thus, this paper, while also presenting several novel mathematical results, is mainly meant to provide a road map through the different neuronally emulated temporal asymmetrical learning rules and their behavior to provide some guidance for possible applications.     

Selectivity in social and asocial learning: investigating the prevalence,effect and development of young children's learning preferences

Culture evolution requires both modification and faithful replication of behaviour, thus it is essential to understand how individuals choose between social and asocial learning. In a quasi-experimental design, 3- and 5-year-olds (176), and adults (52) were presented individually with two novel artificial fruits, and told of the apparatus'' relative difficulty (easy versus hard). Participants were asked if they wanted to attempt the task themselves or watch an experimenter attempt it first; and then had their preference either met or violated. A significant proportion of children and adults (74%) chose to learn socially. For children, this request was efficient, as observing a demonstration made them significantly quicker at the task than learning asocially. However, for 5-year-olds, children who selected asocial learning were also found to be highly efficient at the task, showing that by 5 years children are selective in choosing a learning strategy that is effective for them. Adults further evidenced this trend, and also showed selectivity based on task difficulty. This is the first study to examine the rates, performance outcomes and developmental trajectory of preferences in asocial and social learning, ultimately informing our understanding of innovation.     

Transfer index and mediational learning in tufted capuchins (Cebus apella)

The transfer index (TI) is a discrimination reversal paradigm that requires the achievement of a given prereversal criterion of accuracy. The mediational learning (ML) paradigm is a modification of the TI procedure that features the presentation of three different reversal conditions designed to assess whether prereversal learning is based on purely associative processes or mediated by the use of a strategy (win-stay/lose-shift). These two paradigms have been used with apes and several Old World monkey species, proving to be effective tools for the comparison of species on the basis of their transfer abilities and the nature of their learning processes. However, among New World monkeys, only the squirrel monkey has been tested. Capuchin (Cebusspp.) adaptability and their mastery in using tools have led to controversial interpretations of their cognitive and learning skills. We evaluated their mode of learning and the transfer of learning using the TI and the ML paradigms. We tested four tufted capuchins (Cebus apella)in a WGTA using a variety of stimulus object pairs. The results show that they possess rather good transfer abilities and one subject showed an associative learning mode. None of the subjects showed evidence of learning mediated by a win-stay/lose-shift strategy.     

Investigating the molecular mechanisms of learning and memory using Caenorhabditis elegans

Learning is an essential biological process for survival since it facilitates behavioural plasticity in response to environmental changes. This process is mediated by a wide variety of genes, mostly expressed in the nervous system. Many studies have extensively explored the molecular and cellular mechanisms underlying learning and memory. This review will focus on the advances gained through the study of the nematode Caenorhabditis elegans. C. elegans provides an excellent system to study learning because of its genetic tractability, in addition to its invariant, compact nervous system (~300 neurons) that is well-characterised at the structural level. Importantly, despite its compact nature, the nematode nervous system possesses a high level of conservation with mammalian systems. These features allow the study of genes within specific sensory-, inter- and motor neurons, facilitating the interrogation of signalling pathways that mediate learning via defined neural circuits. This review will detail how learning and memory can be studied in C. elegans through behavioural paradigms that target distinct sensory modalities. We will also summarise recent studies describing mechanisms through which key molecular and cellular pathways are proposed to affect associative and non-associative forms of learning.

     

Learning strategies in matching to sample: if-then and configural learning by pigeons

Pigeons learned a matching-to-sample task with a split training-set design in which half of the stimulus displays were untrained and tested following acquisition. Transfer to the untrained displays along with no novel-stimulus transfer indicated that these pigeons learned the task (partially) via if-then rules. Comparisons to other performance measures indicated that they also partially learned the task via configural learning (learning the gestalt of the whole stimulus display). Differences in the FR-sample requirement (1 vs. 20) had no systematic effect on the type of learning or level of learning obtained. Differences from a previous study [Wright, A.A., 1997. Concept learning and learning strategies. Psychol. Sci. 8, 119-123] are discussed, including the effect of displaying the stimuli vertically (traditional display orientation) or horizontally from the floor.     

Machine learning for epigenetics and future medical applications

Understanding epigenetic processes holds immense promise for medical applications. Advances in Machine Learning (ML) are critical to realize this promise. Previous studies used epigenetic data sets associated with the germline transmission of epigenetic transgenerational inheritance of disease and novel ML approaches to predict genome-wide locations of critical epimutations. A combination of Active Learning (ACL) and Imbalanced Class Learning (ICL) was used to address past problems with ML to develop a more efficient feature selection process and address the imbalance problem in all genomic data sets. The power of this novel ML approach and our ability to predict epigenetic phenomena and associated disease is suggested. The current approach requires extensive computation of features over the genome. A promising new approach is to introduce Deep Learning (DL) for the generation and simultaneous computation of novel genomic features tuned to the classification task. This approach can be used with any genomic or biological data set applied to medicine. The application of molecular epigenetic data in advanced machine learning analysis to medicine is the focus of this review.     

Learning and coding of concepts in neural networks

Our environment consists of virtually an infinite number of scenarios in which we have to function. In order to respond properly to an incoming stimulus, the brain has first to analyze it, and to find out the basic familiar elements that are part of it. In other words, by using a library which contains a relatively small number of basic concepts, the brain analyzes the multitude of incoming events. Some of those basic concepts are innate, but many of them must be learned, in order to accommodate for the arbitrary environment around us. A classifying box is defined as the neural network that finds out the familiar concepts that are present in an incoming stimulus. Models for classifying boxes are introduced, and possible mechanisms by which they may establish their libraries of concepts are suggested, and then compared and evaluated by computer simulations.     

Success-biased social learning: Cultural and evolutionary dynamics

Success bias is a social learning strategy whereby learners tend to acquire the cultural variants of successful individuals. I develop a general model of success-biased social learning for discrete cultural traits with stochastic payoffs, and investigate its dynamics when only two variants are present. I find that success bias inherently favors rare variants, and consequently performs worse than unbiased imitation (i.e. random copying) when success payoffs are at least mildly stochastic and the optimal variant is common. Because of this weakness, success bias fails to replace unbiased imitation in an evolutionary model when selection is fairly weak or when the environment is relatively stable, and sometimes fails to invade at all. I briefly discuss the optimal strength of success bias, the complicated nature of defining success in social learning contexts, and the value of variant frequency as an important source of information to social learners. I conclude with predictions regarding the prevalence of success bias in different behavioral domains.     

Route and landmark learning by rats searching for food

Many foraging animals rely on visual landmarks and/or habitual paths to locate important resources. We examined the degree to which rats rely on these cues when they predicted conflicting food locations. In this foraging task, rats were required to find food which could be located using either a fixed route or a nearby visual landmark. In tests, we found that their subsequent search-based estimates of the food location were the same when animals had acquired a long-term memory of the route, the landmark, or both. We show that the degree to which animals rely on the cues depends not only on the discrepancy between the two cues, but also on whether animals can match the testing “view” with a learned “view” that has been acquired during training.     

Developing a contextualised blended learning framework to enhance medical physics student learning and engagement

As a specialist field of study, medical physicists require a broad range of knowledge and skills to operate competently in their workplace. In Australia, these competencies are accredited by the Australasian College of Physical Scientists and Engineers in Medicine (ACPSEM). Education and training for medical physicists therefore consists of an exhaustive range of knowledge areas. This is made even more challenging due to the extremely diverse backgrounds of students in these specialist courses of study. These factors frequently lead to a disengagement by students with learning activities.To address some of these challenges, the Radiotherapy Physics unit in a Masters level Medical Physics course of study was re-designed to increase active learning that included scaffolded in-class and online tasks and supported by virtual reality simulations. These re-design initiatives were informed by a diverse team including academic and clinical medical physicists as well as education experts.A survey, conducted over two consecutive years was used to gain students perceptions about the re-design. The questions were designed to see if the students felt engaged with the various learning activities.Analysis of the survey data indicates that there was an overall improvement in students’ engagement with the learning activities and the learning content.The paper further discusses nuanced understanding about the ways in which students engaged with the various learning activities including online, in-class, practicals and industry attachments. The paper discusses the appropriately informed learning activities that can be used to improve student engagement for highly specialised, content heavy areas of study.     

Complementary roles of basal ganglia and cerebellum in learning and motor control

The classical notion that the basal ganglia and the cerebellum are dedicated to motor control has been challenged by the accumulation of evidence revealing their involvement in non-motor, cognitive functions. From a computational viewpoint, it has been suggested that the cerebellum, the basal ganglia, and the cerebral cortex are specialized for different types of learning: namely, supervised learning, reinforcement learning and unsupervised learning, respectively. This idea of learning-oriented specialization is helpful in understanding the complementary roles of the basal ganglia and the cerebellum in motor control and cognitive functions.     

Enhancing theoretical understanding of a practical biology course using active and self-directed learning strategies

Laboratories are recognised as central in science education, allowing students to consolidate knowledge and master practical skills, however, their effectiveness has been questioned. Whilst laboratory practicals are useful for students’ learning of basic procedures, they have been shown to be less effective for developing conceptual understanding of the subject. Interactive lectures and bespoke digital resources were utilised in order to enhance theoretical understanding of laboratory practical molecular sessions, thus enabling students to take responsibility for and direct their own learning, encouraging inquiry-based learning. Providing easy to access additional learning resources offered students an opportunity to better prepare themselves for the laboratory, and consolidate their knowledge through subsequent review and self-testing in their own time. Grades before and after implementation of these active learning strategies were analysed to look at the impact on student learning and this study demonstrates that integrating these into a challenging practical biology course improved grades significantly with a concomitant increase in the number of ‘A’ grades attained. Feedback to evaluate use and perceptions of both interactive lectures and digital resources were also analysed. It has been shown here that these activities enhanced student experience and understanding of the course.     

Three models of song learning: Evidence from behavior

Research on avian song learning has traditionally been based on an instructional model, as exemplified by the sensorimotor model of song development. Several large-scale, species-wide field studies of learned birdsongs have revealed that variation is narrowly restricted to certain aspects of song structure. Other aspects are sufficiently stereotyped and so widely shared by species' members that they qualify as species-specific universals. The limitations on natural song variation are difficult to reconcile with a fully open, instructive model of song learning. An alternative model based on memorization by selection postulates a system of innate neural templates that facilitate the recognition and rapid memorization of conspecific song patterns. Behavioral evidence compatible with this model includes learning preferences, rapid conspecific song learning, and widespread ocurrence of species-specific song universals that are recognized innately but fail to develop in songs of social isolates. A third model combines instruction, in the memorization phase, with selection during song production. An overproduced repertoire of plastic songs previously memorized by instruction is winnowed by selection imposed during social interactions at the time of adult song crystallization. Selection during production is well established as a factor in the song development of several species, in the form of action-based learning. The possible role of selective processes in song memorization merits further neurobiological investigation. © 1997 John Wiley & Sons, Inc. J Neurobiol 33: 501–516, 1997     

Extendable and explainable deep learning for pan-cancer radiogenomics research

Radiogenomics is a field where medical images and genomic profiles are jointly analyzed to answer critical clinical questions. Specifically, people want to identify non-invasive imaging biomarkers that are associated with both genomic features and clinical outcomes. Deep learning is an advanced computer science technique that has been applied in many fields, including medical image and genomic data analysis. This review summarizes the current state of deep learning in pan-cancer radiogenomic research, discusses its limitations, and indicates the potential future directions. Traditional machine learning in radiomics, genomics, and radiogenomics have also been briefly discussed. We also summarize the main pan-cancer radiogenomic research resources. Two characteristics of deep learning are emphasized when discussing its application to pan-cancer radiogenomics, which are extendibility and explainability.     

In silico,in vitro,and in vivo machine learning in synthetic biology and metabolic engineering

Among the main learning methods reviewed in this study and used in synthetic biology and metabolic engineering are supervised learning, reinforcement and active learning, and in vitro or in vivo learning.In the context of biosynthesis, supervised machine learning is being exploited to predict biological sequence activities, predict structures and engineer sequences, and optimize culture conditions.Active and reinforcement learning methods use training sets acquired through an iterative process generally involving experimental measurements. They are applied to design, engineer, and optimize metabolic pathways and bioprocesses.The nascent but promising developments with in vitro and in vivo learning comprise molecular circuits performing simple tasks such as pattern recognition and classification.     

A robust deep attention dense convolutional neural network for plant leaf disease identification and classification from smart phone captured real world images

Plant diseases cause significant food loss and hence economic loss around the globe. Therefore, automatic plant disease identification is a primary task to take proper medications for controlling the spread of the diseases. Large variety of plants species and their dissimilar phytopathological symptoms call for the implementation of supervised machine learning techniques for efficient and reliable disease identification and classification. With the development of deep learning strategies, convolutional neural network (CNN) has paved its way for classification of multiple plant diseases by extracting rich features. However, several characteristics of the input images especially captured in real world environment, viz. complex or indistinguishable background, presence of multiple leaves with the diseased leaf, small lesion area, solemnly affect the robustness and accuracy of the CNN modules. Available strategies usually applied standard CNN architectures on the images captured in the laboratory environment and very few have considered practical in-field leaf images for their studies. However, those studies are limited with very limited number of plant species. Therefore, there is need of a robust CNN module which can successfully recognize and classify the dissimilar leaf health conditions of non-identical plants from the in-field RGB images. To achieve the above goal, an attention dense learning (ADL) mechanism is proposed in this article by merging mixed sigmoid attention learning with the basic dense learning process of deep CNN. The basic dense learning process derives new features at higher layer considering all lower layer features and that provides fast and efficient training process. Further, the attention learning process amplifies the learning ability of the dense block by discriminating the meaningful lesion portions of the images from the background areas. Other than adding an extra layer for attention learning, in the proposed ADL block the output features from higher layer dense learning are used as an attention mask to the lower layers. For an effective and fast classification process, five ADL blocks are stacked to build a new CNN architecture named DADCNN-5 for obtaining classification robustness and higher testing accuracy. Initially, the proposed DADCNN-5 module is applied on publicly available extended PlantVillage dataset to classify 38 different health conditions of 14 plant species from 54,305 images. Classification accuracy of 99.93% proves that the proposed CNN module can be used for successful leaf disease identification. Further, the efficacy of the DADCNN-5 model is checked after performing stringent experiments on a new real world plant leaf database, created by the authors. The new leaf database contains 10,851 real-world RGB leaf images of 17 plant species for classifying their 44 distinguished health conditions. Experimental outcomes reveal that the proposed DADCNN-5 outperforms the existing machine learning and standard CNN architectures, and achieved 97.33% accuracy. The obtained sensitivity, specificity and false positive rate values are 96.57%, 99.94% and 0.063% respectively. The module takes approximately 3235 min for training process and achieves 99.86% of training accuracy. Visualization of Class activation mapping (CAM) depicts that DADCNN-5 is able to learn distinguishable features from semantically important regions (i.e. lesion regions) on the leaves. Further, the robustness of the DADCNN-5 is established after experimenting with augmented and noise contaminated images of the practical database.     

PBL与LBL教学法在内科学查房教学中的临床对比研究

目的:比较以问题为基础(PBL)和以授课为基础(LBL)的教学法在内科学查房教学中的应用效果。方法:选取2014年6月-2015年6月内科78名本科实习生参与研究,将学生随机分成PBL教学组(n=39名)和LBL教学组(n=39名),经过为期一年的学习后通过综合考试及问卷调查评价教学效果。结果:PBL组学生的理论考试成绩总分及病例分析题得分显著高于LBL组学生,差异有统计学意义(P0.05);PBL教学组学生在提高学生文献检索的能力、自学能力、解决问题的能力、提高学习效率、锻炼人际沟通能力、培养团队合作精神、有利于理解记忆知识、接受该教学方法的占比明显高于与LBL教学组学生,具有统计学意义(P0.05)。结论:与LBL教学法相比,PBL教学法在教学查房中具有较好的教学效果,在提高学生的文献检索能力、自学能力、解决问题的能力、人际沟通能力、团队合作等综合素质有优势。     

虎皮鹦鹉的认知表现及关联

许多研究认为,不同个体的认知能力存在差异,同一个体的多种认知表现之间可能存在关联。然而,目前对动物认知表现关联的研究结果存在诸多矛盾和争议。本研究以虎皮鹦鹉（Melopsittacus undulatus）为研究对象,通过自我控制、联想学习和反转学习三项认知实验,探究三项表现的联系。实验结果发现,在虎皮鹦鹉中,联想学习表现越好,反转学习亦表现越好。个体对实验装置的新异恐惧（对新环境或新事物的恐惧）越弱,自我控制表现越好。雌、雄个体在三项认知实验中的表现没有显著差异。综上所述,虎皮鹦鹉的一些认知能力存在关联,这暗示了虎皮鹦鹉中可能存在一般认知能力（“g”）。新异恐惧与自我控制表现的强烈相关,说明了勇敢程度可能会影响认知表现。     

Learning set formation in a group of Guinea baboons (Papio papio): Importance of the behavioral variable measured

A recent report of one-trial learning in a group of saddle-back tamarins (Saguinus fuscicollis) conflicted with views of learning set formation based on traditional techniques employing isolated subjects. An experiment is reported in which a group of Guinea baboons (Papio papio) was given a series of discrimination and habit reversal tasks. Both one-trial learning and learning set formation occurred. Analysis of spatial behavior revealed that the group learned in a single trial to discriminate stocked from unstocked zones. Improvement in successes (reinforced “digging”) was progressive but not rapid, indicating learning set formation. It appears that outcomes depend on the behavioral variables chosen.     

Artificial intelligence and machine learning for medical imaging: A technology review

Artificial intelligence (AI) has recently become a very popular buzzword, as a consequence of disruptive technical advances and impressive experimental results, notably in the field of image analysis and processing. In medicine, specialties where images are central, like radiology, pathology or oncology, have seized the opportunity and considerable efforts in research and development have been deployed to transfer the potential of AI to clinical applications. With AI becoming a more mainstream tool for typical medical imaging analysis tasks, such as diagnosis, segmentation, or classification, the key for a safe and efficient use of clinical AI applications relies, in part, on informed practitioners. The aim of this review is to present the basic technological pillars of AI, together with the state-of-the-art machine learning methods and their application to medical imaging. In addition, we discuss the new trends and future research directions. This will help the reader to understand how AI methods are now becoming an ubiquitous tool in any medical image analysis workflow and pave the way for the clinical implementation of AI-based solutions.