Unsupervised Anomaly Detection in Time Series with Recurrent Neural Networks

Total Page:16

File Type:pdf, Size:1020Kb

Unsupervised Anomaly Detection in Time Series with Recurrent Neural Networks DEGREE PROJECT IN TECHNOLOGY, FIRST CYCLE, 15 CREDITS STOCKHOLM, SWEDEN 2019 Unsupervised anomaly detection in time series with recurrent neural networks JOSEF HADDAD CARL PIEHL KTH ROYAL INSTITUTE OF TECHNOLOGY SCHOOL OF ELECTRICAL ENGINEERING AND COMPUTER SCIENCE Unsupervised anomaly detection in time series with recurrent neural networks JOSEF HADDAD, CARL PIEHL Bachelor in Computer Science Date: June 7, 2019 Supervisor: Pawel Herman Examiner: Örjan Ekeberg School of Electrical Engineering and Computer Science Swedish title: Oövervakad avvikelsedetektion i tidsserier med neurala nätverk iii Abstract Artificial neural networks (ANN) have been successfully applied to a wide range of problems. However, most of the ANN-based models do not attempt to model the brain in detail, but there are still some models that do. An example of a biologically constrained ANN is Hierarchical Temporal Memory (HTM). This study applies HTM and Long Short-Term Memory (LSTM) to anomaly detection problems in time series in order to compare their performance for this task. The shape of the anomalies are restricted to point anomalies and the time series are univariate. Pre-existing implementations that utilise these networks for unsupervised anomaly detection in time series are used in this study. We primarily use our own synthetic data sets in order to discover the networks’ robustness to noise and how they compare to each other regarding different characteristics in the time series. Our results shows that both networks can handle noisy time series and the difference in performance regarding noise robustness is not significant for the time series used in the study. LSTM out- performs HTM in detecting point anomalies on our synthetic time series with sine curve trend but a conclusion about the overall best performing network among these two remains inconclusive. iv Sammanfattning Artificiella neurala nätverk (ANN) har tillämpats på många problem. Däremot försöker inte de flesta ANN-modeller efterlikna hjärnan i detalj. Ett exempel på ett ANN som är begränsat till att efterlikna hjärnan är Hierarchical Temporal Memory (HTM). Denna studie tillämpar HTM och Long Short-Term Memory (LSTM) på avvikelsedetektionsproblem i tidsserier för att undersöka vilka styrkor och svagheter de har för detta problem. Avvikelserna i denna studie är begränsade till punktavvikelser och tidsserierna är i endast en variabel. Redan existerande implementationer som utnyttjar dessa nätverk för oövervakad avvikelsedetek- tionsproblem i tidsserier används i denna studie. Vi använder främst våra egna syntetiska tidsserier för att undersöka hur nätverken hanterar brus och hur de hanterar olika egenskaper som en tidsserie kan ha. Våra resultat visar att båda nätverken kan hantera brus och prestationsskillnaden rörande brusrobusthet var inte tillräckligt stor för att urskilja modellerna. LSTM presterade bättre än HTM på att upptäcka punktavvikelser i våra syntetiska tidsserier som följer en sinuskurva men en slutsats angående vilket nätverk som presterar bäst överlag är fortfarande oavgjord. Contents 1 Introduction 1 1.1 Aims and Research Question . .2 1.2 Scope . .3 1.3 Outline . .3 2 Background 4 2.1 Time Series and Anomalies . .4 2.2 HTM . .5 2.2.1 HTM Neuron . .6 2.2.2 HTM Network . .7 2.2.3 HTM activation and learning . .8 2.3 ANNs . .9 2.3.1 Training . 10 2.3.2 RNNs . 10 2.3.3 LSTM . 11 2.4 Related Work . 12 3 Method 14 3.1 HTM Configuration . 14 3.1.1 Network structure . 15 3.1.2 Training . 15 3.1.3 Anomaly labelling . 16 3.2 LSTM configuration . 16 3.2.1 Network structure . 17 3.2.2 Training . 17 3.2.3 Anomaly labelling . 18 3.3 Differences between the used models . 19 3.4 Evaluation and Performance metrics . 19 3.5 Data sets used . 20 v vi CONTENTS 3.5.1 Data sets for testing noise robustness . 20 3.5.2 Data sets for testing time series characteristics . 22 3.5.3 Real-world time series used . 24 4 Results 26 4.1 Noise robustness . 26 4.1.1 Statistical hypothesis testing . 29 4.2 Trend/characteristics results . 29 4.2.1 Statistical hypothesis testing . 34 4.3 Results from real world data sets . 35 4.3.1 Occupancy t4013 . 35 4.3.2 Ec2_request_latency_system_failure . 37 5 Discussion 39 5.1 Noise robustness . 39 5.2 Time series and anomaly characteristics . 40 5.3 Real world time series performance . 41 5.4 General comparison . 42 5.5 Limitations . 42 5.6 Biological approach . 44 5.7 Future Work . 44 6 Conclusions 46 Bibliography 47 A Time series graphs 51 A.1 Noise robustness graphs . 51 A.2 Synthetic time series graphs . 53 Acronyms ANN Artificial Neural Network. BPTT Back-Propagation Through Time. HTM Hierarchical Temporal Memory. LSTM Long-Short Term Memory. NAB Numenta Anomaly Benchmark. RNN Recurrent Neural Network. SDR Sparse Distributed Representation. vii Chapter 1 Introduction Advances in neuroscience have allowed for a greatly increased understanding of the structure and function of different parts of the brain. Simultaneously, great advancements have been made in the field of machine learning. Artifi- cial Neural Networks (ANN), in particular, have been of great interest in the research community and have been successfully applied to a wide range of problems, from medical diagnosis [1] to playing games [2]. However, most ANN-based models do not attempt to model the brain in any detail [3]. These advancements are mostly driven by mathematically derived models devised to perform specific tasks which do not utilise our increased understanding of the brain. They often also require extensive training in order to perform these tasks and cannot easily be generalised to perform other tasks. An alternative approach is to use biologically inspired models by trying to mimic the way the human brain processes information. There are a few examples of brain inspired methods, one of the more detailed ones is the Hierarchical Temporal Memory (HTM). HTM is an evolving attempt to model the structure and function of the neocortex first introduced by Hawkins [4]. The model is based on Mount- castle’s [5] proposal that all the regions of the neocortex, which makes up roughly 80% of the brain and is responsible for higher-order functions such as cognition and language, follow a similar neuroanatomical design [5]. It is hy- pothesised that the difference in the functionality by different regions mainly arises out of different inputs. This implies, in theory, that a faithful model of the neocortex could be trained to perform multifarious function that would be considered as the backbone of intelligent behaviour. Hawkins claims that the neocortex achieves this by memorising patterns and constantly making predic- tions based on those memories. This way, the neocortex can learn spatial and 1 2 CHAPTER 1. INTRODUCTION temporal patterns in its environment. When input is received predictions are made based on learned patterns. Due to this ability to make inference based on temporal patterns, it would be of interest to test the performance of HTM in time series anomaly detection tasks. Time series data, data captured over a period of time, occurs naturally in many real world scenarios and analysis of time series data has been of in- terest in fields such as engineering, economics and medicine [6]. Patterns in a time series data that deviate from expected or normal behaviour are considered to be anomalies [6]. This means that a tool which can accurately predict the future values in a time series can also be used as an anomaly detector. Time series anomaly detection can be useful in many areas such as sleep monitoring [6], jet engine operation [7] and intrusion detection for computer networks [8]. Anomalies can be difficult to detect because it can be difficult to determine if a pattern in a time series is considered to be normal on not since an anomaly in one process can be considered normal behaviour in another [6]. An identifica- tion of an anomalous behaviour can be done with simple threshold heuristics [9], but those often require knowledge of the data sets and implementation by a human with deep domain knowledge. The major challenge lies in capturing dependencies among multiple variables as well as identifying long-term and short-term repeating patterns. This is where traditional approaches, such as auto regressive methods, can fall short [10]. An alternative approach, which has seen increasing popularity as of late, is using ANNs as anomaly detectors. Different types of ANNs have been applied to a wide range of time series prob- lems, such as predicting flour prices and modelling the amount of littering in the North Sea [6]. 1.1 Aims and Research Question The aim of this thesis is to evaluate a HTM network in the task of detecting point anomalies in time series data. The strength and weaknesses of the net- work are compared to a state-of-the-art ANN-based approach. The selected ANN is the Long-Short Term Memory (LSTM) Recurrent Neural Network (RNN). RNNs are well suited for time series data since they can represent in- formation from an arbitrarily long context window [11]. However, they have traditionally been difficult to train and perform worse with very long-term temporal dependencies [12]. Adding LSTM units to these networks has been shown to remedy some of these issues, allowing the networks to achieve state- of-the-art performance in time series anomaly detection tasks [6][13]. CHAPTER 1. INTRODUCTION 3 The research question for this thesis is: How does HTM compare to LSTM in time series point anomaly detection tasks? 1.2 Scope HTM and LSTM are compared when performing unsupervised anomaly de- tection in single variable streaming data. In particular, this study focuses on robustness to noise and ability to recognise anomalies in time series with dif- ferent characteristics.
Recommended publications
  • Memory-Prediction Framework for Pattern
    Memory–Prediction Framework for Pattern Recognition: Performance and Suitability of the Bayesian Model of Visual Cortex Saulius J. Garalevicius Department of Computer and Information Sciences, Temple University Room 303, Wachman Hall, 1805 N. Broad St., Philadelphia, PA, 19122, USA [email protected] Abstract The neocortex learns sequences of patterns by storing This paper explores an inferential system for recognizing them in an invariant form in a hierarchical neural network. visual patterns. The system is inspired by a recent memory- It recalls the patterns auto-associatively when given only prediction theory and models the high-level architecture of partial or distorted inputs. The structure of stored invariant the human neocortex. The paper describes the hierarchical representations captures the important relationships in the architecture and recognition performance of this Bayesian world, independent of the details. The primary function of model. A number of possibilities are analyzed for bringing the neocortex is to make predictions by comparing the the model closer to the theory, making it uniform, scalable, knowledge of the invariant structure with the most recent less biased and able to learn a larger variety of images and observed details. their transformations. The effect of these modifications on The regions in the hierarchy are connected by multiple recognition accuracy is explored. We identify and discuss a number of both conceptual and practical challenges to the feedforward and feedback connections. Prediction requires Bayesian approach as well as missing details in the theory a comparison between what is happening (feedforward) that are needed to design a scalable and universal model. and what you expect to happen (feedback).
    [Show full text]
  • Neuromorphic Architecture for the Hierarchical Temporal Memory Abdullah M
    IEEE TRANSACTIONS ON EMERGING TOPICS IN COMPUTATIONAL INTELLIGENCE 1 Neuromorphic Architecture for the Hierarchical Temporal Memory Abdullah M. Zyarah, Student Member, IEEE, Dhireesha Kudithipudi, Senior Member, IEEE, Neuromorphic AI Laboratory, Rochester Institute of Technology Abstract—A biomimetic machine intelligence algorithm, that recognition and classification [3]–[5], prediction [6], natural holds promise in creating invariant representations of spatiotem- language processing, and anomaly detection [7], [8]. At a poral input streams is the hierarchical temporal memory (HTM). higher abstraction, HTM is basically a memory based system This unsupervised online algorithm has been demonstrated on several machine-learning tasks, including anomaly detection. that can be trained on a sequence of events that vary over time. Significant effort has been made in formalizing and applying the In the algorithmic model, this is achieved using two core units, HTM algorithm to different classes of problems. There are few spatial pooler (SP) and temporal memory (TM), called cortical early explorations of the HTM hardware architecture, especially learning algorithm (CLA). The SP is responsible for trans- for the earlier version of the spatial pooler of HTM algorithm. forming the input data into sparse distributed representation In this article, we present a full-scale HTM architecture for both spatial pooler and temporal memory. Synthetic synapse design is (SDR) with fixed sparsity, whereas the TM learns sequences proposed to address the potential and dynamic interconnections and makes predictions [9]. occurring during learning. The architecture is interweaved with A few research groups have implemented the first generation parallel cells and columns that enable high processing speed for Bayesian HTM. Kenneth et al., in 2007, implemented the the HTM.
    [Show full text]
  • Memory-Prediction Framework for Pattern Recognition: Performance
    Memory–Prediction Framework for Pattern Recognition: Performance and Suitability of the Bayesian Model of Visual Cortex Saulius J. Garalevicius Department of Computer and Information Sciences, Temple University Room 303, Wachman Hall, 1805 N. Broad St., Philadelphia, PA, 19122, USA [email protected] Abstract The neocortex learns sequences of patterns by storing This paper explores an inferential system for recognizing them in an invariant form in a hierarchical neural network. visual patterns. The system is inspired by a recent memory- It recalls the patterns auto-associatively when given only prediction theory and models the high-level architecture of partial or distorted inputs. The structure of stored invariant the human neocortex. The paper describes the hierarchical representations captures the important relationships in the architecture and recognition performance of this Bayesian world, independent of the details. The primary function of model. A number of possibilities are analyzed for bringing the neocortex is to make predictions by comparing the the model closer to the theory, making it uniform, scalable, knowledge of the invariant structure with the most recent less biased and able to learn a larger variety of images and observed details. their transformations. The effect of these modifications on The regions in the hierarchy are connected by multiple recognition accuracy is explored. We identify and discuss a number of both conceptual and practical challenges to the feedforward and feedback connections. Prediction requires Bayesian approach as well as missing details in the theory a comparison between what is happening (feedforward) that are needed to design a scalable and universal model. and what you expect to happen (feedback).
    [Show full text]
  • The Neuroscience of Human Intelligence Differences
    Edinburgh Research Explorer The neuroscience of human intelligence differences Citation for published version: Deary, IJ, Penke, L & Johnson, W 2010, 'The neuroscience of human intelligence differences', Nature Reviews Neuroscience, vol. 11, pp. 201-211. https://doi.org/10.1038/nrn2793 Digital Object Identifier (DOI): 10.1038/nrn2793 Link: Link to publication record in Edinburgh Research Explorer Document Version: Peer reviewed version Published In: Nature Reviews Neuroscience Publisher Rights Statement: This is an author's accepted manuscript of the following article: Deary, I. J., Penke, L. & Johnson, W. (2010), "The neuroscience of human intelligence differences", in Nature Reviews Neuroscience 11, p. 201-211. The final publication is available at http://dx.doi.org/10.1038/nrn2793 General rights Copyright for the publications made accessible via the Edinburgh Research Explorer is retained by the author(s) and / or other copyright owners and it is a condition of accessing these publications that users recognise and abide by the legal requirements associated with these rights. Take down policy The University of Edinburgh has made every reasonable effort to ensure that Edinburgh Research Explorer content complies with UK legislation. If you believe that the public display of this file breaches copyright please contact [email protected] providing details, and we will remove access to the work immediately and investigate your claim. Download date: 02. Oct. 2021 Nature Reviews Neuroscience in press The neuroscience of human intelligence differences Ian J. Deary*, Lars Penke* and Wendy Johnson* *Centre for Cognitive Ageing and Cognitive Epidemiology, Department of Psychology, University of Edinburgh, Edinburgh EH4 2EE, Scotland, UK. All authors contributed equally to the work.
    [Show full text]
  • On Intelligence As Memory
    Artificial Intelligence 169 (2005) 181–183 www.elsevier.com/locate/artint Book review Jeff Hawkins and Sandra Blakeslee, On Intelligence, Times Books, 2004. On intelligence as memory Jerome A. Feldman International Computer Science Institute, Berkeley, CA 94704-1198, USA Available online 3 November 2005 On Intelligence by Jeff Hawkins with Sandra Blakeslee has been inspirational for non- scientists as well as some of our most distinguished biologists, as can be seen from the web site (http://www.onintelligence.org). The book is engagingly written as a first person memoir of one computer engineer’s search for enlightenment on how human intelligence is computed by our brains. The central insight is important—much of our intelligence comes from the ability to recognize complex situations and to predict their possible outcomes. There is something fundamental about the brain and neural computation that makes us intelligent and AI should be studying it. Hawkins actually understates the power of human associative memory. Because of the massive parallelism and connectivity, the brain essentially reconfigures itself to be constantly sensitive to the current context and goals [1]. For example, when you are planning to buy some kind of car, you start noticing them. The book is surely right that better AI systems would follow if we could develop programs that were more like human memory. For whatever reason, memory as such is no longer studied much in AI—the Russell and Norvig [3] text has one index item for memory and that refers to semantics. From a scientific AI/Cognitive Science perspective, the book fails to tackle most of the questions of interest.
    [Show full text]
  • Mapping Mind-Brain Development: Towards a Comprehensive Theory
    Journal of Intelligence Review Mapping Mind-Brain Development: Towards a Comprehensive Theory George Spanoudis 1,* and Andreas Demetriou 2,3 1 Psychology Department, University of Cyprus, 1678 Nicosia, Cyprus 2 Department of Psychology, University of Nicosia, 1700 Nicosia, Cyprus; [email protected] 3 Cyprus Academy of Science, Letters, and Arts, 1700 Nicosia, Cyprus * Correspondence: [email protected]; Tel.: +357-22892969 Received: 20 January 2020; Accepted: 20 April 2020; Published: 26 April 2020 Abstract: The relations between the developing mind and developing brain are explored. We outline a theory of intellectual development postulating that the mind comprises four systems of processes (domain-specific, attention and working memory, reasoning, and cognizance) developing in four cycles (episodic, realistic, rule-based, and principle-based representations, emerging at birth, 2, 6, and 11 years, respectively), with two phases in each. Changes in reasoning relate to processing efficiency in the first phase and working memory in the second phase. Awareness of mental processes is recycled with the changes in each cycle and drives their integration into the representational unit of the next cycle. Brain research shows that each type of processes is served by specialized brain networks. Domain-specific processes are rooted in sensory cortices; working memory processes are mainly rooted in hippocampal, parietal, and prefrontal cortices; abstraction and alignment processes are rooted in parietal, frontal, and prefrontal and medial cortices. Information entering these networks is available to awareness processes. Brain networks change along the four cycles, in precision, connectivity, and brain rhythms. Principles of mind-brain interaction are discussed. Keywords: brain; mind; development; intelligence; mental and brain changes; brain networks 1.
    [Show full text]
  • Intelligence As a Developing Function: a Neuroconstructivist Approach
    Journal of Intelligence Article Intelligence as a Developing Function: A Neuroconstructivist Approach Luca Rinaldi 1,2,* and Annette Karmiloff-Smith 3 1 Department of Brain and Behavioural Sciences, University of Pavia, Pavia 27100, Italy 2 Milan Center for Neuroscience, Milano 20126, Italy 3 Centre for Brain and Cognitive Development, Birkbeck, London WC1E 7HX, UK; [email protected] * Correspondence: [email protected]; Tel.: +39-02-6448-3775 Academic Editors: Andreas Demetriou and George Spanoudis Received: 23 December 2016; Accepted: 27 April 2017; Published: 29 April 2017 Abstract: The concept of intelligence encompasses the mental abilities necessary to survival and advancement in any environmental context. Attempts to grasp this multifaceted concept through a relatively simple operationalization have fostered the notion that individual differences in intelligence can often be expressed by a single score. This predominant position has contributed to expect intelligence profiles to remain substantially stable over the course of ontogenetic development and, more generally, across the life-span. These tendencies, however, are biased by the still limited number of empirical reports taking a developmental perspective on intelligence. Viewing intelligence as a dynamic concept, indeed, implies the need to identify full developmental trajectories, to assess how genes, brain, cognition, and environment interact with each other. In the present paper, we describe how a neuroconstructivist approach better explains why intelligence can rise or fall over development, as a result of a fluctuating interaction between the developing system itself and the environmental factors involved at different times across ontogenesis. Keywords: intelligence; individual differences; development; neuroconstructivism; emergent structure; developmental trajectory 1.
    [Show full text]
  • Conversation with Jeff Hawkins – on Defining Intelligence
    Episode 12: Conversation with Jeff Hawkins – On Defining Intelligence Matt: 00:00 Today on the Numenta On Intelligence podcast. Jeff: 00:04 You see something to believe the earth is flat and you give them all these facts that don't fit that model. Now, the Earth is flat predicts certain things. Right. And, and, and they will do their damnedest to try to fit these new facts into their model, no matter how, that's what they're going to, otherwise you have to throw the whole thing away and start over again, which is a really disconcerting thing for them. Matt: 00:26 It is. It is uncomfortable, to have to throw away a whole frame of reference. Jeff: 00:30 Basically, the models you build in the world are your reality. Matt: 00:34 It's your belief system. Jeff: 00:34 It's your belief system. It's your reality. This is what you know, that's the, oh, that's what it boils down to. There is no other reality in terms of your head. This is it. Instead of, if you say, well, you know what? Everything you believed these references, and you have to start over again. Um, then it's kind of like, oh, you're going back to square one. Matt: 00:49 Jeff Hawkins on defining intelligence. What is it? What makes us intelligent? How was our intelligence different from a mouse'? We're narrowing down the answers to some pretty existential questions. I'm Numenta community manager, Matt Taylor.
    [Show full text]
  • Hierarchical Temporal Memory (HTM)
    Hierarchical Temporal Memory (HTM) Computational Cognitive Neuroscience COSC 521, Spring 2019 Corey Johnson Resources for information on HTM Numenta.org HTM school NuPIC Jeff Hawkins 2004 book: On Intelligence: How a New Understanding of the Brain will Lead to the Creation of Truly Intelligent Machines (Note: some slide graphics are from Numenta.org, HTM school) 2 Hierarchical ● Higher levels → abstraction & permanence Temporal ● Change over time: patterns Memory ● Sparse Distributed Representation (SDR) 3 HTMs ● Biologically plausible model for intelligence ● Based on pyramidal neurons ● Neocortex micro-columns as building blocks "The neocortex comprises about 75% of the volume of the human brain and it is the seat of most of what we think of as intelligence." -Jeff Hawkins 4 HTMs continued ● Can learn / recall / infer high-order sequences ● Local learning rules, no global supervisor (HW) ● Relies on sparse distributed representation (SDR) ○ Fault tolerance ○ High capacity ● Learns by modeling the growth of new synapses 5 ANN Pyramidal HTM 6 HTM Layers (Lect. 2 slides) 7 Sparse Distributed Representation Fundamental to HTM systems SDR: 'Language of intelligence' Sparse as opposed to dense binary code, many bits are needed SDR bit similarity → similar semantic meaning Semantic error is key to generalization Cell only needs a few connections to neighbor to match pattern 8 Applications Good for: ● Data streams that change over time: text, GPS, dates, numbers ● Data with inherent structure ● System where many models are required rather than
    [Show full text]
  • From Neural Networks to Deep Learning: Zeroing in on the Human Brain
    From Neural Networks to Deep Learning: Zeroing in on the Human Brain Pondering the brain with the help of machine learning expert Andrew Ng and researcher- turned-author-turned-entrepreneur Jeff Hawkins. By Jonathan Laserson DOI: 10.1145/2000775.2000787 n the spring of 1958, David Hubble and Torsten Wiesel stood helpless in their lab at John Hopkins University. In front of them lay an anesthetized cat. In the cat’s skull, just above its visual cortex (an area in the back of the brain), there was a 3mm-wide hole through which electrodes recorded the response of several neurons. Hubble and Wiesel had Iconjectured that these neurons would fire when shown the right stimulus on the retina. If only they had any idea what that stimulus could be. For days they had shown the poor cat different shapes and shiny spots of light, testing different areas of the retina, but nothing elicited a response. In their book, Brain and the Visual Perception, Hubble and Wiesel give a personal account of this experience: The break came one long day in which example of what is now known as “orien- using the following experiment. They we held onto one cell for hour after hour. tation selective cells,” a type of cell which collected a bunch of natural images— To find a region of retina from which our is actually prevalent in the visual cortex. trees, lakes, leaves, and so on—and spots gave any hint of responses took These cells fire when they detect an edge extracted thousands of small image many hours, but we finally found a place oriented at a particular angle, in a specif- patches, 16x16 pixels each, from ran- that gave vague hints of responses.
    [Show full text]
  • On the Prospects for Building a Working Model of the Visual Cortex
    On the Prospects for Building a Working Model of the Visual Cortex Thomas Dean Glenn Carroll and Richard Washington Brown University, Providence, RI Google Inc., Mountain View, CA [email protected] {gcarroll,rwashington}@google.com Abstract new or recycled? How important is the role of Moore’s law in our pursuit of human-level perception? The full story Human visual capability has remained largely beyond the delves into the role of time, hierarchy, abstraction, complex- reach of engineered systems despite intensive study and con- siderable progress in problem understanding, algorithms and ity, symbolic reasoning, and unsupervised learning. It owes computing power. We posit that significant progress can be much to insights that have been discovered and forgotten at made by combining existing technologies from computer vi- least once every other generation for many decades. sion, ideas from theoretical neuroscience and the availability Since J.J. Gibson (1950) presented his theory of ecologi- of large-scale computing power for experimentation. From cal optics, scientists have followed his lead by trying to ex- a theoretical standpoint, our primary point of departure from plain perception in terms of the invariants that organisms current practice is our reliance on exploiting time in order to learn. Peter Foldi¨ ak´ (1991) and more recently Wiskott and turn an otherwise intractable unsupervised problem into a lo- Sejnowski (2002) suggest that we learn invariances from cally semi-supervised, and plausibly tractable, learning prob- lem. From a pragmatic perspective, our system architecture temporal input sequences by exploiting the fact that sensory follows what we know of cortical neuroanatomy and provides input tends to vary quickly while the environment we wish a solid foundation for scalable hierarchical inference.
    [Show full text]
  • Fluid and Flexible Minds: Intelligence Reflects Synchrony in the Brain's Intrinsic Network Architecture
    RESEARCH Fluid and flexible minds: Intelligence reflects synchrony in the brain’s intrinsic network architecture ∗ ∗ Michael A. Ferguson1,2 , Jeffrey S. Anderson2, and R. Nathan Spreng1 1Laboratory of Brain and Cognition, Human Neuroscience Institute, Department of Human Development, Cornell University, Ithaca, NY, 14853 2Departments of Bioengineering and Neuroradiology, University of Utah, Salt Lake City, UT, 84132 Keywords: intelligence, fMRI, resting state functional connectivity, machine learning, cognition an open access journal ABSTRACT Human intelligence has been conceptualized as a complex system of dissociable cognitive processes, yet studies investigating the neural basis of intelligence have typically emphasized the contributions of discrete brain regions or, more recently, of specific networks of functionally connected regions. Here we take a broader, systems perspective in order to investigate whether intelligence is an emergent property of synchrony within the brain’s intrinsic network architecture. Using a large sample of resting-state fMRI and cognitive data (n = 830), we report that the synchrony of functional interactions within and across distributed brain networks reliably predicts fluid and flexible intellectual functioning. By adopting a whole-brain, systems-level approach, we were able to reliably predict individual differences in human intelligence by characterizing features of the brain’s intrinsic network architecture. These findings hold promise for the eventual development of neural markers to Citation: Ferguson,
    [Show full text]