In this paper, we present MATRICS, a humanmachine hybrid system that accurately performs geopolitical forecasting by combining crowdsourcing with ensemble machine learning on online data. The system employs a pair of parallel, but highly-interconnected processing pipelines to perform “machine-aided human forecasting” and “human-aided machine forecasting”. This configuration allows the machine to provide information to the human population, saving research time and reducing fatigue, while simultaneously allowing the human population to provide feedback to the machine learning components, allowing them to filter data sources and quickly adapt to a task via online machine learning. The final forecast for each question was computed as an aggregate of the human and machine responses. The system was evaluated using data collected during the IARPA Hybrid Forecasting Competition, in which it answered 187 forecasting questions with a mean Brier score of 0.27 using volunteers and participants that were recruited via Amazon Mechanical Turk and open-source “big” data scraped from online sources such as social media, search engine results, and online historical data.

We provide here a novel method, called hypercolumn sparsification, to achieve high recognition performance for convolutional neural networks (CNNs) despite low-precision weights and activities during both training and test phases. This method is applicable to any CNN architecture that operates on signal patterns (e.g., audio, image, video) to extract information such as class membership. It operates on the stack of feature maps in each of the cascading feature matching and pooling layers through the processing hierarchy of the CNN by an explicit competitive process ( k -WTA, winner take all) that generates a sparse feature vector at each spatial location. This principle is inspired by local brain circuits, where neurons tuned to respond to different patterns in the incoming signals from an upstream region inhibit each other using interneurons, such that only the ones that are maximally activated survive the quenching threshold. We show this process of sparsification is critical for probabilistic learning of low-precision weights and bias terms, thereby making pattern recognition amenable for energy-efficient hardware implementations. Further, we show that hypercolumn sparsification could lead to more data-efficient learning as well as having an emergent property of significantly pruning down the number of connections in the network. A theoretical account and empirical analysis are provided to understand these effects better.

Research investigating the dynamics of coupled physical systems has demonstrated that small feedback delays can allow a dynamic response system to anticipate chaotic behavior. This counterintuitive phenomenon, termed anticipatory synchronization, has been observed in coupled electrical circuits, laser semi-conductors, and artificial neurons. Recent research indicates that the same process might also support the ability of humans to anticipate the occurrence of chaotic behavior in other individuals. Motivated by this latter work, the current study examined whether the process of feedback delay induced anticipatory synchronization could be employed to develop an interactive artificial agent capable of anticipating chaotic human movement. Results revealed that incorporating such delays within the movement-control dynamics of an artificial agent not only enhances an artificial agent's ability to anticipate chaotic human behavior, but to synchronize with such behavior in a manner similar to natural human-human anticipatory synchronization. The implication of these findings for the development of human-machine interaction systems is discussed.

Here we describe a spiking neural network model for obtaining the conditional probability between two random variables from the synaptic weight between two corresponding neurons. Our method does not aim to mimic biologically plausible processes in the brain, but instead aims to replace slow Bayesian updating algorithms with highly parallel, simple computations by taking inspiration from Spike Timing Dependent Plasticity (STDP), a naturally occurring phenomena of neuronal synapses. The neuronal network we develop can operate in two different modes to both learn the joint distribution of variables, and read out conditional probabilities based on partial input. Furthermore, it relies on integer-valued neurons and synapses, making it amenable to hardware-based implementations. We demonstrate this capability of our novel probabilistic computation unit by encoding synthetically generated input data streams into spike trains and letting the synaptic weight of interest converge to the correct conditional probability.

By integrating theories and methodologies from a diverse range of scientific disciplines (e.g., physics, neuroscience, cognitive science, psychology and robotics engineering) the present work is aimed at harnessing self-organized anticipatory synchronization in order to advance humanrobotic interaction (HRI). This phenomenon is characterized by the emergence of anticipatory behavior by one system coupled to the chaotic behavior of another, following the introduction of short self-referential delays in the coordinating system. The current set of studies involved the creation of an artificial agent based on a time-delayed, low-dimensional dynamical model capable of behaving prospectively during an interaction with a human actor performing complex, unpredictable behaviors. By achieving characteristics similar to those observed during natural human interaction and coordination, the time-delayed modeling approached advocated here provides the potential for considerable future advancements in HRI.

With increased rates of smartphone theft over the past decade, mobile authentication systems that operate on a continual basis are a necessity to meet increasing demands for user privacy, device usage, and authentication accuracy. Rather than forcing end users to continually self-authenticate via password pins or through other means on a time-interval basis [2-7], a system that continuously authenticates users provides a more frictionless relationship between a user's device and its physical security. Such a system, if effectively operating on a low-powered, unobtrusive, and secure basis, would make it viable for most consumer mobile devices. In this paper, we build upon our work in [1] to provide a novel authentication scheme that meets these requirements for a commonly adopted system. Our system, iSentinel, hopes to provide an unobtrusive, low-powered solution for detecting and responding to common theft scenarios by continuously authenticating mobile devices in use cases such as walking, texting, and driving.

Two experiments are reported showing that behavior exhibited in manual tracking is consistent with behavior predicted by a dynamical systems phenomenon known as anticipating synchronization (Voss, 2000). They extend a prior investigation of the effect of delay on anticipatory manual tracking (Stepp, 2009) by also manipulating coupling strength. The coupling scheme in Experiment 1 and that in Experiment 2 go beyond the single delayed feedback coupling used in previous research and articulations of anticipating synchronization. These advanced coupling arrangements are addressed using an extended formulation which allows for multiple feedback delays, a continuous range of delay, or even coupling to real future values. The latter case is specifically investigated in Experiment 2, which utilizes a navigation task that provides a natural way to speak about coupling to future values.

Finding the balance between security, privacy, and usability for mobile authentication has been an active area of research for the past several years. Many researchers have taken advantage of the availability of multiple sensors on mobile devices and have used these data to train classifiers to authenticate users. For example, implicit authentication algorithms have been developed based on behavior patterns identified from a combination of sensors including location, co-location, application usage, biometric measurements, continuity of interaction between the user and the phone, and possession of the phone [1,2,3,4,5]. Furthermore, the onboard sensors of mobile devices have previously been used to identify users based on touch [6] and fusions of touch and speech inputs [7]. However, a system utilizing low-power onboard electronics for anomaly detection and user classification is lacking. Here, we report on the performance of two subsystems tested in a controlled use scenario to classify and authenticate users of a mobile device. The overall system utilizes two subsystems for anomaly detection and user classification: (1) a neuromorphic chip for continuous, low-power, online monitoring and classification, and (2) an early warning system (EWS) algorithm for longer duration time-series behavioral and biometric classification.

We propose that feedback-delayed manual tracking performance is limited by fundamental constraints imposed by the physics of negative group delay. To test this hypothesis, the results of an experiment in which subjects demonstrate both reactive and predictive dynamics are modeled by a linear system with delay-induced negative group delay. Although one of the simplest real-time predictors conceivable, this model explains key components of experimental observations. Most notably, it explains the observation that prediction time linearly increases with feedback delay, up to a certain point when tracking performance deteriorates. It also explains the transition from reactive to predictive behavior with increasing feedback delay. The model contains only one free parameter, the feedback gain, which has been fixed by comparison with one set of experimental observations for the reactive case. Our model provides quantitative predictions that can be tested in further experiments.

We present an end-to-end image-processing pipeline that is accelerated using coupled spin-torque oscillator (STO) arrays for its key computational kernels. Coupled oscillator arrays are used throughout the pipeline, including the computation of spectral and spatial transforms in the saliency stages and convolution in the CNN-based classifiers in the back end. We present results for two types of oscillator models, a generic, parameterized, model that captures variations in coupling asymmetry, locking range, and noise, and a hardware calibrated spice model of a coupled STO array with data curve fitted to a closed-form C++ model. The pipeline was run over three benchmark datasets, the Neovision2 Tower and Helicopter datasets [1] and the DARPA Vivid dataset [2]. We show the sensitivity of the output to parameter variations in the generic model and show equivalent performance for the STO model.

During rest, the mammalian cortex displays spontaneous neural activity. Spiking of single neurons during rest has been described as irregular and asynchronous. In contrast, recent in vivo and in vitro population measures of spontaneous activity, using the LFP, EEG, MEG or fMRI suggest that the default state of the cortex is critical, manifested by spontaneous, scale-invariant, cascades of activity known as neuronal avalanches. Criticality keeps a network poised for optimal information processing, but this view seems to be difficult to reconcile with apparently irregular single neuron spiking. Here, we simulate a 10,000 neuron, deterministic, plastic network of spiking neurons. We show that a combination of short- and long-term synaptic plasticity enables these networks to exhibit criticality in the face of intrinsic, i.e. self-sustained, asynchronous spiking. Brief external perturbations lead to adaptive, long-term modification of intrinsic network connectivity through long-term excitatory plasticity, whereas long-term inhibitory plasticity enables rapid self-tuning of the network back to a critical state. The critical state is characterized by a branching parameter oscillating around unity, a critical exponent close to -3/2 and a long tail distribution of a self-similarity parameter between 0.5 and 1.

Neuromorphic hardware are designed by drawing inspiration from biology to overcome limitations of current computer architectures while forging the development of a new class of autonomous systems that can exhibit adaptive behaviors. Several designs in the recent past are capable of emulating large scale networks but avoid complexity in network dynamics by minimizing the number of dynamic variables that are supported and tunable in hardware. We believe that this is due to the lack of a clear understanding of how to design self-tuning complex systems. It has been widely demonstrated that criticality appears to be the default state of the brain and manifests in the form of spontaneous scale-invariant cascades of neural activity. Experiment, theory and recent models have shown that neuronal networks at criticality demonstrate optimal information transfer, learning and information processing capabilities that affect behavior. In this perspective article, we argue that understanding how large scale neuromorphic electronics can be designed to enable emergent adaptive behavior will require an understanding of how networks emulated by such hardware can self-tune local parameters to maintain criticality as a set-point. We believe that such capability will enable the design of truly scalable intelligent systems using neuromorphic hardware that embrace complexity in network dynamics rather than avoiding it.

In J. J. Gibson's classic paper "The Problem of Temporal Order in Stimulation and Perception" (1966a), he referred to the difficulties encountered when attempting a sharp distinction between memory and perception as "the muddle of memory." Resolution of the muddle by J. J. Gibson proceeded by blurring the distinction itself. We develop the conjugate "muddle of anticipation" similarly by blurring the sharp distinction traditionally drawn between anticipation and perception. The subsequent redefinition of the problem is grounded in strong anticipation equated with anticipating synchronization-that which arises from a system itself via lawful regularities embedded in the system's ordinary mode of function. We identify the fit of strong anticipation's properties to J. J. Gibson's ecological approach and in so doing introduce the possibility of a potentially deep connection between them, namely, that the coordination of perception with surroundings (direct perception) is a special case of strong anticipation.

Self-organization as a concept has appeared in several arenas, including explanations of physical phenomena, biological systems, and intelligence. A guiding principle for self-organizing systems, especially at the level of intelligent systems, has not been settled upon. So-called autocatakinetic (ACK) systems attempt to provide such a principle through macroscopic thermodynamics but to date have not been formally defined. We attempt to extend ACK by developing a formal model commensurate with its defining properties.

When a person standing upright raises an arm on cue, muscles of the left and right sides of the body exhibit changes prior to and specific to the responding arm. We had standing participants perform a visual lexical decision task ("is this letter string a word?"), responding yes by raising one arm and no by raising the other arm. We recorded onset of the arm movement and onset of electromyographic activity in thigh, trunk, and shoulder muscles. We observed the expected responding arm specificity and found that the onset difference favoring word decisions was evident in similar magnitude at all measurement sites, with the difference at the levels of thigh, trunk and shoulder muscles available 225, 189, and 120 ms, respectively, prior to its manifestation at the level of arm movement. We discuss including (a) whole body reaction time along with event-related potentials in determining the decision-response, brain-body temporal relation and (b) response execution along with response initiation in investigating mental chronometry. (C) 2011 Elsevier Ireland Ltd. All rights reserved.

Cognitive science has always included multiple methodologies and theoretical commitments. The philosophy of cognitive science should embrace, or at least acknowledge, this diversity. Bechtel's (2009a) proposed philosophy of cognitive science, however, applies only to representationalist and mechanist cognitive science, ignoring the substantial minority of dynamically oriented cognitive scientists. As an example of nonrepresentational, dynamical cognitive science, we describe strong anticipation as a model for circadian systems (Stepp & Turvey, 2009). We then propose a philosophy of science appropriate to nonrepresentational, dynamical cognitive science.