General-purpose Learning Research Articles

Synergistic information supports modality integration and flexible learning in neural networks solving multiple tasks.

Striking progress has been made in understanding cognition by analyzing how the brain is engaged in different modes of information processing. For instance, so-called synergistic information (information encoded by a set of neurons but not by any subset) plays a key role in areas of the human brain linked with complex cognition. However, two questions remain unanswered: (a) how and why a cognitive system can become highly synergistic; and (b) how informational states map onto artificial neural networks in various learning modes. Here we employ an information-decomposition framework to investigate neural networks performing cognitive tasks. Our results show that synergy increases as networks learn multiple diverse tasks, and that in tasks requiring integration of multiple sources, performance critically relies on synergistic neurons. Overall, our results suggest that synergy is used to combine information from multiple modalities-and more generally for flexible and efficient learning. These findings reveal new ways of investigating how and why learning systems employ specific information-processing strategies, and support the principle that the capacity for general-purpose learning critically relies on the system's information dynamics.

High-efficiency reinforcement learning with hybrid architecture photonic integrated circuit

Reinforcement learning (RL) stands as one of the three fundamental paradigms within machine learning and has made a substantial leap to build general-purpose learning systems. However, using traditional electrical computers to simulate agent-environment interactions in RL models consumes tremendous computing resources, posing a significant challenge to the efficiency of RL. Here, we propose a universal framework that utilizes a photonic integrated circuit (PIC) to simulate the interactions in RL for improving the algorithm efficiency. High parallelism and precision on-chip optical interaction calculations are implemented with the assistance of link calibration in the hybrid architecture PIC. By introducing similarity information into the reward function of the RL model, PIC-RL successfully accomplishes perovskite materials synthesis task within a 3472-dimensional state space, resulting in a notable 56% improvement in efficiency. Our results validate the effectiveness of simulating RL algorithm interactions on the PIC platform, highlighting its potential to boost computing power in large-scale and sophisticated RL tasks.

A Graph-Based Neural Approach to Linear Sum Assignment Problems.

Linear assignment problems are well-known combinatorial optimization problems involving domains such as logistics, robotics and telecommunications. In general, obtaining an optimal solution to such problems is computationally infeasible even in small settings, so heuristic algorithms are often used to find near-optimal solutions. In order to attain the right assignment permutation, this study investigates a general-purpose learning strategy that uses a bipartite graph to describe the problem structure and a message passing Graph Neural Network (GNN) model to learn the correct mapping. Comparing the proposed structure with two existing DNN solutions, simulation results show that the proposed approach significantly improves classification accuracy, proving to be very efficient in terms of processing time and memory requirements, due to its inherent parameter sharing capability. Among the many practical uses that require solving allocation problems in everyday scenarios, we decided to apply the proposed approach to address the scheduling of electric smart meters access within an electricity distribution smart grid infrastructure, since near-real-time energy monitoring is a key element of the green transition that has become increasingly important in recent times. The results obtained show that the proposed graph-based solver, although sub-optimal, exhibits the highest scalability, compared with other state-of-the-art heuristic approaches. To foster the reproducibility of the results, we made the code available at https://github.com/aircarlo/GNN_LSAP.

An Advance Study of an Efficient CNN-Grounded Deep Learning Classification Technique for the Diagnosis of IoT based Cardiac Arrhythmias

Deep Learning, or DL for short, is an emerging subfield within the larger discipline of machine learning in today's world. The study being conducted in this area is progressing at an immediate stride, and the discoveries are contributing to the progression of technology. Deep learning (DL) methods were developed with the intention of developing a general-purpose learning method that would enable the gradual learning of characteristics at multiple levels without relying on human-engineered features. This was the goal of deep learning. Because of this, the system is able to acquire intricate purposes and directly map input to output by making use of the data that it has acquired which is based on Internet of things (IoTs). This study places an emphasis on the application of CNN (Convolutional Neural Networks), which are a subcategory of DNN (Deep Neural Networks), and it develops an efficient layered CNN for the classification of ECG arrhythmias. Even while FC-ANNs (Fully Connected Artificial Neural Networks), which are sometimes referred to as Multilayer-Perceptron networks, are effective in categorising ECG arrhythmias, the optimization process for many classification networks takes a significant amount of time in terms of computation. In addition, the features extracted by engineers are what define the accuracy of the categorization of ECG arrhythmias. An improved CNN based filtering, feature abstraction, and classification prototypical is established in order to conduct an accurate analysis of an electrocardiogram (ECG). When measured against ANN, the performance was found to have an accuracy rating of 99.6%. Consequently, the CNN model that was suggested is useful to doctors in arriving at the definitive diagnosis of AFL (atrial flutter), AFIB (atrial fibrillation), VFL (ventricular flutter), and VT (ventricular tachycardia). It includes denoising, feature extraction, and categorization as part of its functionality.

Social and asocial learning in zebrafish are encoded by a shared brain network that is differentially modulated by local activation

Group living animals use social and asocial cues to predict the presence of reward or punishment in the environment through associative learning. The degree to which social and asocial learning share the same mechanisms is still a matter of debate. We have used a classical conditioning paradigm in zebrafish, in which a social (fish image) or an asocial (circle image) conditioned stimulus (CS) have been paired with an unconditioned stimulus (US=food), and we have used the expression of the immediate early gene c-fos to map the neural circuits associated with each learning type. Our results show that the learning performance is similar to social and asocial CSs. However, the brain regions activated in each learning type are distinct and a community analysis of brain network data reveals segregated functional submodules, which seem to be associated with different cognitive functions involved in the learning tasks. These results suggest that, despite localized differences in brain activity between social and asocial learning, they share a common learning module and social learning also recruits a specific social stimulus integration module. Therefore, our results support the occurrence of a common general-purpose learning module, that is differentially modulated by localized activation in social and asocial learning.

Artificial nervous systems – a technology to achieve biologically modeled intelligence and control for robotics

Migrating from machine learning and deep learning into the next wave of technology will likely require biological replication rather than biological inspiration. An approach to achieving this requires duplicating entire nervous systems, or at least parts thereof. In theory, these artificial nervous systems (ANS) could reproduce everything required for a system to be biologically intelligent even to the point of being self-aware. This would additionally entail that the resultant systems have the ability to acquire information from both their internal and external environments as well as having the ability to act within the external environment using locomotion and manipulators. Robots are a natural answer for the resultant mechanism and if supplied with an artificial nervous system, the robot might be expected to achieve biologically modelled intelligence (BMI) and control. This paper will provide an overview of the tools for creating artificial nervous systems, as well as provide a roadmap for utilizing the tools to develop robots with general-purpose learning skills and biologically modelled intelligence.

Developmental Network-2: The Autonomous Generation of Optimal Internal-Representation Hierarchy.

It is very challenging for machine learning methods to reach the goal of general-purpose learning since there are so many complicated situations in different tasks. The learning methods need to generate flexible internal representations for all scenarios met before. The hierarchical internal representation is considered as an efficient way to build such flexible representations. By hierarchy, we mean important local features in the input can be combined to form higher level features with more context. In this work, we analyze how our proposed general-purpose learning framework-the developmental network-2 (DN-2)-autonomously generates internal hierarchy with new mechanisms. Specifically, DN-2 incrementally allocates neuronal resources to different levels of representation during learning instead of handcrafting static boundaries among different levels of representation. We present the mathematical proof to demonstrate that optimal properties in terms of maximum likelihood (ML) are established under the conditions of limited learning experience and resources. The phoneme recognition and real-world visual navigation experiments that are of different modalities and include many different situations are designed to investigate general-purpose learning capability of DN-2. The experimental results show that DN-2 successfully learns different tasks. The formed internal hierarchical representations focus on important features, and the invariant abstract arise from optimal internal representations. We believe that DN-2 is in the right way toward fully autonomous learning.

CNN based deep learning methods for precise analysis of cardiac arrhythmias

In contemporary day, Deep Learning (DL) is a developing discipline in the science of Machine Learning (ML) (ML). The research in this field is evolving extremely fast and its consequence leads to breakthrough in advance technology. Deep learning approaches are meant to gradually learn characteristics from several layers by adopting a general purpose learning mechanism, without relying on the human built features. This enables the system to learn the complicated functions and translate the input to the output straight from the data. This study effort primarily focuses on emphasising the Convolutional Neural Networks (CNNs), a kind of Deep Neural Networks (DNNs) and develop an 11 layered CNN for effective ECG arrhythmia classification. In this study the relevance of CNNs, the major building blocks and layers are explored, the design of the suggested CNN model is described.

Robust, source-independent biases in children's use of socially and individually acquired information.

Culture has an extraordinary influence on human behavior, unparalleled in other species. Some theories propose that humans possess learning mechanisms biologically selected specifically for social learning, which function to promote rapid enculturation. If true, it follows that information acquired via observation of another's activity might be responded to differently, compared with equivalent information acquired through one's own exploration, and that this should be the case in even very young children. To investigate this, we compared children's responses to information acquired either socially or from personal experience. The task we used allowed direct comparison between these alternative information sources, as the information value was equivalent across conditions, which has not been true of previous methods used to tackle similar questions. Across two 18-month- to 5-year-old samples (recruited in the United Kingdom and China), we found that children performed similarly following information acquired from social demonstrations, compared with personal experience. Children's use of the information thus appeared independent of source. Furthermore, children's suboptimal performance showed evidence of a consistent bias driven by motivation for exploration as well as exploitation, which was apparent across both conditions and in both samples. Our results are consistent with the view that apparent peculiarities identified in human social information use could be developmental outcomes of general-purpose learning and motivational biases, as opposed to mechanisms that have been biologically selected specifically for the acquisition of cultural information. (PsycInfo Database Record (c) 2021 APA, all rights reserved).

Content-based image retrieval system for HRCT lung images: assisting radiologists in self-learning and diagnosis of Interstitial Lung Diseases

Content-based Image Retrieval (CBIR) is a technique that can exploit the wealth of the data stored in a repository and help radiologists in decision making by providing references to the image in hand. A CBIR system for High-Resolution Computed Tomography (HRCT) lung images depicting signs of Interstitial Lung Diseases (ILDs) can be built and used as a self-learning tool for budding radiologists. The study of a few lung image retrieval systems available in the literature identifies some important issues that need to be taken care of. In most of the works, the creation of the reference database involves painstaking manual activity, which is time-consuming and needs skilled labor. A lot of human interventions are required, particularly for the proper delineation of the region of interest (ROI) that represents pathology in each of the images in a database. In most cases, the size of the ROIs representing different disease findings are fixed (i.e., either a fixed size square or circle), which at times may not be a proper representation of the disease pattern and as a consequence, it might limit the system’s performance. Until date, a few learning-based approaches have been developed for content-based image retrieval of HRCT lung images, which either learn the similarity using a classifier or get trained through relevance feedback. For medical image analysis, the availability of labelled data for learning makes these learning-based retrieval systems meaningful as it enhances their performance in contrast to their simple distance-based counterpart. The objective of this paper is to develop a CBIR system for ILDs that is reliable and needs minimal human intervention. The paper evaluates the performance of three popular segmentation algorithms. It identifies the best for the effective and automated delineation of an arbitrary region of interest (AROI) depicting the sign of ILDs on HRCT images of the thorax in contrast to the manual delineation of fixed size ROI. This minimizes the manual effort for the creation and maintenance of the reference database, as well as the manual delineation of AROI during query formation. Moreover, AROI created through the automated clustering is found to have a better representation of disease patterns. Three recently proposed general-purpose learning based CBIR techniques are implemented and tested for retrieval of HRCT lung images depicting the sign of ILDs. The best method is suggested after careful evaluation of all the competing techniques.

Deep Learning for Period Classification of Historical Hebrew Texts

In this study, we address the interesting task of classifying historical texts by their assumed period of writ-ing. This task is useful in digital humanity studies where many texts have unidentified publication dates.For years, the typical approach for temporal text classification was supervised using machine-learningalgorithms. These algorithms require careful feature engineering and considerable domain expertise todesign a feature extractor to transform the raw text into a feature vector from which the classifier couldlearn to classify any unseen valid input. Recently, deep learning has produced extremely promising re-sults for various tasks in natural language processing (NLP). The primary advantage of deep learning isthat human engineers did not design the feature layers, but the features were extrapolated from data witha general-purpose learning procedure. We investigated deep learning models for period classification ofhistorical texts. We compared three common models: paragraph vectors, convolutional neural networks (CNN) and recurrent neural networks (RNN), and conventional machine-learning methods. We demon-strate that the CNN and RNN models outperformed the paragraph vector model and the conventionalsupervised machine-learning algorithms. In addition, we constructed word embeddings for each timeperiod and analyzed semantic changes of word meanings over time.

Grandmaster level in StarCraft II using multi-agent reinforcement learning.

Many real-world applications require artificial agents to compete and coordinate with other agents in complex environments. As a stepping stone to this goal, the domain of StarCraft has emerged as an important challenge for artificial intelligence research, owing to its iconic and enduring status among the most difficult professional esports and its relevance to the real world in terms of its raw complexity and multi-agent challenges. Over the course of a decade and numerous competitions1-3, the strongest agents have simplified important aspects of the game, utilized superhuman capabilities, or employed hand-crafted sub-systems4. Despite these advantages, no previous agent has come close to matching the overall skill of top StarCraft players. We chose to address the challenge of StarCraft using general-purpose learning methods that are in principle applicable to other complex domains: a multi-agent reinforcement learning algorithm that uses data from both human and agent games within a diverse league of continually adapting strategies and counter-strategies, each represented by deep neural networks5,6. We evaluated our agent, AlphaStar, in the full game of StarCraft II, through a series of online games against human players. AlphaStar was rated at Grandmaster level for all three StarCraft races and above 99.8% of officially ranked human players.

Intelligent fault diagnosis method of planetary gearboxes based on convolution neural network and discrete wavelet transform

Considering the planetary gearbox vibration signals show highly non-stationary and non-linear behavior because of wind turbines (WTs) often working under time-varying running conditions, we propose an effective and reliable method based on convolution neural network (CNN) and discrete wavelet transformation (DWT) to identify the fault conditions of planetary gearboxes. Firstly, the discrete wavelet transformation is used with the goal of presenting more salient and comprehensive time-frequency distributed representation. Secondly, the deep hierarchical structure of CNN constructed by the alternating convolution layers and subsample layers is trained using a forward transmitting rule of greedy training layer by layer and translates the low-level features of input to the high-level features in order to identify the internal characteristic. Finally, a top classifier Softmax is added at uppermost layer of CNN and the backpropagation process is conducted to fine-tune the parameters of CNN, establishing the mapping relation among the feature space and the fault space. Thus, feature extraction process and fault recognition are incorporated into a general-purpose learning procedure. The experimental results of the fault diagnosis for the planetary gearbox demonstrated the effectiveness and feasibility of the proposed method.

An Overview of Neural Network

Neural networks represent a brain metaphor for information processing. These models are biologically inspired rather than an exact replica of how the brain actually functions. Neural networks have been shown to be very promising systems in many forecasting applications and business classification applications due to their ability to learn from the data. This article aims to provide a brief overview of artificial neural network. The artificial neural network learns by updating the network architecture and connection weights so that the network can efficiently perform a task. It can learn either from available training patterns or automatically learn from examples or input-output relations. Neural network-based models continue to achieve impressive results on longstanding machine learning problems, but establishing their capacity to reason about abstract concepts has proven difficult. Building on previous efforts to solve this important feature of general-purpose learning systems, our latest paper sets out an approach for measuring abstract reasoning in learning machines, and reveals some important insights about the nature of generalization itself. Artificial neural networks can learn by example like the way humans do. An artificial neural net is configured for a specific application like pattern recognition through a learning process. Learning in biological systems consists of adjustments to the synaptic connections that exist between neurons. This is true of artificial neural networks as well. Artificial neural networks can be applied to an increasing number of real-world problems of considerable complexity. They are used for solving problems that are too complex for conventional technologies or those types of problems that do not have an algorithmic solution.

Learning Radio Resource Management in RANs: Framework, Opportunities, and Challenges

In the fifth generation (5G) of mobile broadband systems, radio resource management (RRM) will reach unprecedented levels of complexity. To cope with the ever more sophisticated RRM functionalities and the growing variety of scenarios, while carrying out the prompt decisions required in 5G, this manuscript presents a lean RRM architecture that capitalizes on recent advances in the field of machine learning in combination with the large amount of data readily available in the network from measurements and system observations. The architecture consists of a learner (or a few), which learns RRM policies directly from the data gathered in the network using a single general-purpose learning framework, and a set of distributed actors, which execute RRM policies issued by the learner and repeatedly generate samples of experience. Thus, the complexity of RRM is shifted to the design of the learning framework, while the RRM algorithms derived from this framework are executed in a computationally efficient distributed manner at the radio access nodes. The potential of this approach is verified in a pair of pertinent scenarios, and future directions on applications of machine learning to RRM are discussed. Although we focus on a mobile broadband context, the concepts proposed hereafter extend to any radio access network technology where one can conceive the idea of a central learning unit gathering data from distributed actors.

Learning of hierarchical serial patterns emerges in infancy.

Recursive, hierarchically organized serial patterns provide the underlying structure in many cognitive and motor domains including speech, language, music, social interaction, and motor action. We investigated whether learning of hierarchical patterns emerges in infancy by habituating 204 infants to different hierarchical serial patterns and then testing for discrimination and generalization of such patterns. Results indicated that 8- to 10-month-old and 12- to 14-month-old infants exhibited sensitivity to the difference between hierarchical and non-hierarchical structure but that 4- to 6-month-old infants did not. These findings demonstrate that the ability to perceive, learn, and generalize recursive, hierarchical, pattern rules emerges in infancy and add to growing evidence that general-purpose pattern learning mechanisms emerge during the first year of life.

Child first language and adult second language are both tied to general-purpose learning systems

Do the mechanisms underlying language in fact serve general-purpose functions that preexist this uniquely human capacity? To address this contentious and empirically challenging issue, we systematically tested the predictions of a well-studied neurocognitive theory of language motivated by evolutionary principles. Multiple metaanalyses were performed to examine predicted links between language and two general-purpose learning systems, declarative and procedural memory. The results tied lexical abilities to learning only in declarative memory, while grammar was linked to learning in both systems in both child first language and adult second language, in specific ways. In second language learners, grammar was associated with only declarative memory at lower language experience, but with only procedural memory at higher experience. The findings yielded large effect sizes and held consistently across languages, language families, linguistic structures, and tasks, underscoring their reliability and validity. The results, which met the predicted pattern, provide comprehensive evidence that language is tied to general-purpose systems both in children acquiring their native language and adults learning an additional language. Crucially, if language learning relies on these systems, then our extensive knowledge of the systems from animal and human studies may also apply to this domain, leading to predictions that might be unwarranted in the more circumscribed study of language. Thus, by demonstrating a role for these systems in language, the findings simultaneously lay a foundation for potentially important advances in the study of this critical domain.

Road area detection method based on DBNN for robot navigation using single camera in outdoor environments

PurposeThe purpose of this study is to develop a cost-effective autonomous wheelchair robot navigation method that assists the aging population.Design/methodology/approachNavigation in outdoor environments is still a challenging task for an autonomous mobile robot because of the highly unstructured and different characteristics of outdoor environments. This study examines a complete vision guided real-time approach for robot navigation in urban roads based on drivable road area detection by using deep learning. During navigation, the camera takes a snapshot of the road, and the captured image is then converted into an illuminant invariant image. Subsequently, a deep belief neural network considers this image as an input. It extracts additional discriminative abstract features by using general purpose learning procedure for detection. During obstacle avoidance, the robot measures the distance from the obstacle position by using estimated parameters of the calibrated camera, and it performs navigation by avoiding obstacles.FindingsThe developed method is implemented on a wheelchair robot, and it is verified by navigating the wheelchair robot on different types of urban curve roads. Navigation in real environments indicates that the wheelchair robot can move safely from one place to another. The navigation performance of the developed method and a comparison with laser range finder (LRF)-based methods were demonstrated through experiments.Originality/valueThis study develops a cost-effective navigation method by using a single camera. Additionally, it utilizes the advantages of deep learning techniques for robust classification of the drivable road area. It performs better in terms of navigation when compared to LRF-based methods in LRF-denied environments.

Homeostatic Agent for General Environment

Abstract One of the essential aspect in biological agents is dynamic stability. This aspect, called homeostasis, is widely discussed in ethology, neuroscience and during the early stages of artificial intelligence. Ashby’s homeostats are general-purpose learning machines for stabilizing essential variables of the agent in the face of general environments. However, despite their generality, the original homeostats couldn’t be scaled because they searched their parameters randomly. In this paper, first we re-define the objective of homeostats as the maximization of a multi-step survival probability from the view point of sequential decision theory and probabilistic theory. Then we show that this optimization problem can be treated by using reinforcement learning algorithms with special agent architectures and theoretically-derived intrinsic reward functions. Finally we empirically demonstrate that agents with our architecture automatically learn to survive in a given environment, including environments with visual stimuli. Our survival agents can learn to eat food, avoid poison and stabilize essential variables through theoretically-derived single intrinsic reward formulations.

Cell types for our sense of location: where we are and where we are going.

Technological advances in profiling cells along genetic, anatomical and physiological axes have fomented interest in identifying all neuronal cell types. This goal nears completion in specialized circuits such as the retina, while remaining more elusive in higher order cortical regions. We propose that this differential success of cell type identification may not simply reflect technological gaps in co-registering genetic, anatomical and physiological features in the cortex. Rather, we hypothesize it reflects evolutionarily driven differences in the computational principles governing specialized circuits versus more general-purpose learning machines. In this framework, we consider the question of cell types in medial entorhinal cortex (MEC), a region likely to be involved in memory and navigation. While MEC contains subsets of identifiable functionally defined cell types, recent work employing unbiased statistical methods and more diverse tasks reveals unsuspected heterogeneity and adaptivity in MEC firing patterns. This suggests MEC may operate more as a generalist circuit, obeying computational design principles resembling those governing other higher cortical regions.