Incremental Slow Feature Analysis Research Articles

A dual fast and slow feature interaction in biologically inspired visual recognition of human action

Computational neuroscience studies have examined the human visual system through functional magnetic resonance imaging (fMRI) and identified a model where the mammalian brain pursues two independent pathways for recognizing biological movement tasks. On the one hand, the dorsal stream analyzes the motion information by applying optical flow, which considers the fast features. On the other hand, the ventral stream analyzes the form information with slow features. The proposed approach suggests that the motion perception of the human visual system comprises fast and slow feature interactions to identify biological movements. The form features in the visual system follow the application of the active basis model (ABM) with incremental slow feature analysis (IncSFA). Episodic observation is required to extract the slowest features, whereas the fast features update the processing of motion information in every frame. Applying IncSFA provides an opportunity to abstract human actions and use action prototypes. However, the fast features are obtained from the optical flow division, which gives an opportunity to interact with the system as the final recognition is performed through a combination of the optical flow and ABM-IncSFA information and through the application of kernel extreme learning machine. Applying IncSFA into the ventral stream and involving slow and fast features in the recognition mechanism are the major contributions of this research. The two human action datasets for benchmarking (KTH and Weizmann) and the results highlight the promising performance of this approach in model modification.

State Identification of Smart Devices Using Acoustic Fingerprinting

The widespread use of smart phones has brought the security and privacy problems. In the past research, fingerprinting of physical device can be used in authenticating system. However, using the fingerprinting to identify and track the user is useful in many legitimate scenes. In this paper, we propose a novel acoustic fingerprinting approach that uses the microphones and speakers of devices to assess the state of target device. While a person is walking or running with a smart phone, a certain amount of vibration is generated, which can be investigated as the basis of state identification. For mining the subtle vibration variations, firstly, a special high-frequency control simulation to speakers is introduced. Then an unsupervised learning method called Incremental Slow Feature Analysis (IncSFA) is applied to construct the state fingerprinting to represent the slow variation triggered by external driving. At last, two alternate approaches as support vector machine (SVM) and Bayes classifier are used to classify the different states from different smart phones.

Optimal Curiosity-Driven Modular Incremental Slow Feature Analysis.

Consider a self-motivated artificial agent who is exploring a complex environment. Part of the complexity is due to the raw high-dimensional sensory input streams, which the agent needs to make sense of. Such inputs can be compactly encoded through a variety of means; one of these is slow feature analysis (SFA). Slow features encode spatiotemporal regularities, which are information-rich explanatory factors (latent variables) underlying the high-dimensional input streams. In our previous work, we have shown how slow features can be learned incrementally, while the agent explores its world, and modularly, such that different sets of features are learned for different parts of the environment (since a single set of regularities does not explain everything). In what order should the agent explore the different parts of the environment? Following Schmidhuber's theory of artificial curiosity, the agent should always concentrate on the area where it can learn the easiest-to-learn set of features that it has not already learned. We formalize this learning problem and theoretically show that, using our model, called curiosity-driven modular incremental slow feature analysis, the agent on average will learn slow feature representations in order of increasing learning difficulty, under certain mild conditions. We provide experimental results to support the theoretical analysis.

Video anomaly detection using deep incremental slow feature analysis network

Existing anomaly detection (AD) approaches rely on various hand‐crafted representations to represent video data and can be costly. The choice or designing of hand‐crafted representation can be difficult when faced with a new dataset without prior knowledge. Motivated by feature learning, e.g. deep leaning and the ability to directly learn useful representations and model high‐level abstraction from raw data, the authors investigate the possibility of using a universal approach. The objective is learning data‐driven high‐level representation for the task of video AD without relying on hand‐crafted representation. A deep incremental slow feature analysis (D‐IncSFA) network is constructed and applied to directly learning progressively abstract and global high‐level representations from raw data sequence. The D‐IncSFA network has the functionalities of both feature extractor and anomaly detector that make AD completion in one step. The proposed approach can precisely detect global anomaly such as crowd panic. To detect local anomaly, a set of anomaly maps, produced from the network at different scales, is used. The proposed approach is universal and convenient, working well in different types of scenarios with little human intervention and low memory and computational requirements. The advantages are validated by conducting extensive experiments on different challenge datasets.

Continual curiosity-driven skill acquisition from high-dimensional video inputs for humanoid robots

In the absence of external guidance, how can a robot learn to map the many raw pixels of high-dimensional visual inputs to useful action sequences? We propose here Continual Curiosity driven Skill Acquisition (CCSA). CCSA makes robots intrinsically motivated to acquire, store and reuse skills. Previous curiosity-based agents acquired skills by associating intrinsic rewards with world model improvements, and used reinforcement learning to learn how to get these intrinsic rewards. CCSA also does this, but unlike previous implementations, the world model is a set of compact low-dimensional representations of the streams of high-dimensional visual information, which are learned through incremental slow feature analysis. These representations augment the robot's state space with new information about the environment. We show how this information can have a higher-level (compared to pixels) and useful interpretation, for example, if the robot has grasped a cup in its field of view or not. After learning a representation, large intrinsic rewards are given to the robot for performing actions that greatly change the feature output, which has the tendency otherwise to change slowly in time. We show empirically what these actions are (e.g., grasping the cup) and how they can be useful as skills. An acquired skill includes both the learned actions and the learned slow feature representation. Skills are stored and reused to generate new observations, enabling continual acquisition of complex skills. We present results of experiments with an iCub humanoid robot that uses CCSA to incrementally acquire skills to topple, grasp and pick-place a cup, driven by its intrinsic motivation from raw pixel vision.

Value and reward based learning in neurorobots

Organisms are equipped with value systems that signal the salience of environmental cues to their nervous system, causing a change in the nervous system that results in modification of their behavior. These systems are necessary for an organism to adapt its behavior when an important environmental event occurs. A value system constitutes a basic assumption of what is good and bad for an agent. These value systems have been effectively used in robotic systems to shape behavior. For example, many robots have used models of the dopaminergic system to reinforce behavior that leads to rewards. Other modulatory systems that shape behavior are acetylcholine's effect on attention, norepinephrine's effect on vigilance, and serotonin's effect on impulsiveness, mood, and risk. Moreover, hormonal systems such as oxytocin and its effect on trust constitute as a value system. A recent Research Topic in Frontiers of Neurorobotics explored value and reward based learning. The topic comprised of nine papers on research involving neurobiologically inspired robots whose behavior was shaped by value and reward learning, adapted through interaction with the environment, or shaped by extracting value from the environment. Value systems are often linked to reward systems in neurobiology and in modeling. For example, Jayet Bray and her colleagues developed a neurorobotic system that learned to categorize the valence of speech through positive verbal encouragement, much like a baby would (Jayet Bray et al., 2013). Their virtual robot, which interacted with a human partner, was controlled by a large-scale spiking neuron model of the visual cortex, premotor cortex, and reward system. An important issue in both biological and artificial reward systems is the credit assignment problem that is, how can a distal cue be linked to a reward. In other words, how can you extract the stimulus that predicts a future reward from all the noisy stimuli that you are faced with? Soltoggio and colleagues introduce the principle of rare correlations to resolve this issue (Soltosggio et al., 2013). By using Rarely Correlating Hebbian Plasticity, they demonstrated classical and operant conditioning in a set of human-robot experiments with the iCub robot. The notion of value and reward has often been formalized in reinforcement learning systems. For example, Li and colleagues show that reinforcement learning, in the form of a dynamic actor-critic model, can be used to tune central pattern generators in a humanoid robot (Li et al., 2013). Through interaction with the environment, this dynamical system developed biped locomotion on a NAO robot that could adapt its gaits to different conditions. Elfwing and colleagues introduced a scaled version of free-energy reinforcement learning (FERL) and applied it to visual recognition and navigation tasks (Elfwing et al., 2013). This novel algorithm was shown to be significantly better than standard FERL and feedforward neural network RL. Another related method, Linearly solvable Markov Decision Process (LMDP) has been shown to have advantages over RL in optimal control policy (Kinjo et al., 2013). Kinjo and colleagues demonstrated the power of LMDP for robot control by applying the method to a pole balancing task, and a visually guided navigation problem using their Spring Dog robot which has six degrees-of-freedom. Value does need not be reward-based; curiosity, harm, novelty, and uncertainty can all carry a value signal. For example, in a biomimetic model of the cortex, basal ganglia and phasic dopamine, Bolado-Gomez and colleagues (Bolado-Gomez and Gurney, 2013) showed that intrinsically motivated operant learning (i.e., action discovery) could replicate rodent experiments, in a virtual robot. In this case, phasic dopaminergic neuromodulation carried a novelty salience signal, rather than the more conventional reward signal. In a model called CURIOUSity-DRiven, Modular, Incremental Slow Feature Analysis (Curious Dr. MISFA), Luciw and colleagues showed that curiosity could shape the behavior of an iCub robot in a multi-context environment (Luciw et al., 2013). Their model was inspired by cortical regions of the brain involved in unsupervised learning, as well as neuromodulatory systems responsible for providing intrinsic rewards through dopamine and regulating levels of attention through norepinephrine. Different neuromodulatory systems in the brain may be related to different aspects of value (Krichmar, 2013). In a model of multiple neuromodulatory systems, Krichmar showed that interactions between the dopaminergic (reward), serotoninergic (harm aversion), and the cholinergic/noradrenergic (novelty) systems could lead to interesting behavioral control in an autonomous robot. Finally, in an interesting position paper, Friston, Adams, and Montague suggest that value is evidence, specifically log Bayesian evidence (Friston et al., 2012). They propose that reward or cost functions that underlie value in conventional models of optimal control can be cast as prior beliefs about future states, which is simply accumulation of evidence through Bayesian updating of posterior beliefs. As can be gleaned from reading the papers in the Research Topic, as well as the empirical evidence and studies they are built on, Value and Reward Based Learning is an active and broad area of research. The application to neurorobotics is important for several reasons: (1) It provides an embodied platform for testing hypotheses regarding the neural correlates of value and reward, (2) it provides a means to test more theoretical hypotheses on the acquisition of value and its function for biological and artificial systems, and (3) it may lead to the development of improved learning systems in robots and other autonomous agents.

An intrinsic value system for developing multiple invariant representations with incremental slowness learning

Curiosity Driven Modular Incremental Slow Feature Analysis (CD-MISFA;) is a recently introduced model of intrinsically-motivated invariance learning. Artificial curiosity enables the orderly formation of multiple stable sensory representations to simplify the agent's complex sensory input. We discuss computational properties of the CD-MISFA model itself as well as neurophysiological analogs fulfilling similar functional roles. CD-MISFA combines 1. unsupervised representation learning through the slowness principle, 2. generation of an intrinsic reward signal through learning progress of the developing features, and 3. balancing of exploration and exploitation to maximize learning progress and quickly learn multiple feature sets for perceptual simplification. Experimental results on synthetic observations and on the iCub robot show that the intrinsic value system is essential for representation learning. Representations are typically explored and learned in order from least to most costly, as predicted by the theory of curiosity.

Incremental Slow Feature Analysis: Adaptive Low-Complexity Slow Feature Updating from High-Dimensional Input Streams

We introduce here an incremental version of slow feature analysis (IncSFA), combining candid covariance-free incremental principal components analysis (CCIPCA) and covariance-free incremental minor components analysis (CIMCA). IncSFA's feature updating complexity is linear with respect to the input dimensionality, while batch SFA's (BSFA) updating complexity is cubic. IncSFA does not need to store, or even compute, any covariance matrices. The drawback to IncSFA is data efficiency: it does not use each data point as effectively as BSFA. But IncSFA allows SFA to be tractably applied, with just a few parameters, directly on high-dimensional input streams (e.g., visual input of an autonomous agent), while BSFA has to resort to hierarchical receptive-field-based architectures when the input dimension is too high. Further, IncSFA's updates have simple Hebbian and anti-Hebbian forms, extending the biological plausibility of SFA. Experimental results show IncSFA learns the same set of features as BSFA and can handle a few cases where BSFA fails.

1A1-P04 Sparse F-IncSFA for Action Recognition(Communication Robot)

High dimensional input streams and unsupervised learning are two important factors in the area of humanoids and processes of the actions and movements of human. Our Fast Incremental Slow Feature Analysis (F-IncSFA) can learn and extract the few significant features of the complex sensory input sequences regarding high-level spatio-temporal conceptions. In this paper, the application of the F-IncSFA and some of its structure to make a hierarchical compound network made of F-IncSFA has been described. Also the techniques developed by adding efficient sparse coding as an encoder and a preprocessing step before an application of the F-IncSFA. The efficient sparse coding can dramatically reduces the dimension of extracted features and outcome of the efficient sparse coding are quite small as compared with the size of high-dimension video obtained by humanoid or human action. It has revealed excellent and promising dimension reduction by this preprocessor.