Peter Bock

Gray-scale ALIAS

Based on the paradigm of collective learning systems, ALIAS (adaptive learning image analysis system) is an adaptive image-processing engine specifically designed to detect anomalies in otherwise normal images and signals. To accomplish this, ALIAS requires only one pass through a training set, which typically consists of less than 100 samples. The original version of ALIAS

ALISA: Adaptive Learning Image and Signal Analysis

ALISA (Adaptive Learning Image and Signal Analysis) is an adaptive statistical learning engine that may be used to detect and classify the surfaces and boundaries of objects in images. The engine has been designed, implemented, and tested at both the George Washington University and the Research Institute for Applied Knowledge Processing in Ulm, Germany over the last nine years with major funding from Robert Bosch GmbH and Lockheed-Martin Corporation. The design of ALISA was inspired by the multi-path cortical- column architecture and adaptive functions of the mammalian visual cortex.

Identification of Objects-of-Interest in X-Ray Images

The objective of this research is to automatically detect and locate devices-of-interest (DOI) in x-ray images, even if partially obscured by devices of no interest, using a new ALISA Component Module. This preliminary study was performed using a single DOI, a 9mm Colt Beretta, but the solution method can easily accommodate other DOIs. Results obtained in real-time (a few seconds) revealed a robust and accurate classifier that could easily assist security personnel at the defined venue: carry-on luggage x-ray machines in airports. This research project was funded by the defense threat reduction agency (DTRA).

A parallel implementation of collective learning systems theory: Adaptive Learning Image Analysis System (ALIAS)

An alternative to preprogrammed rule-based Artificial Intelligence is a hierarchical network of cellular automata which acquire their knowledge through learning based on a series of trial-and-error interactions with an evaluating Environment, much as humans do. The input to the hierarchical network is provided by a set of sensors which perceive the external world. Based upon this perceived information and past experience (memory), the learning automata synthesize collections of trial responses. Periodically the automata estimate the effectiveness of these collections using either internal evaluations (unsupervised learning) or external evaluations from the Environment (supervised learning), modifying their memories accordingly. Known as Collective Learning Systems Theory, this paradigm has been applied to many sophisticated gaming problems, demonstrating robust learning and dynamic adaptivity. Based on a versatile architecture for massively parallel networks of processors for Collective Learning Systems, a Transputer-based parallel-processing image processing engine comprising 32 learning cells and 32 non-learning cells has been applied to a sophisticated image processing task: the scale-invariant and translation-invariant detection of anomalous features in otherwise “normal” images. In cooperation with Robert Bosch GmbH, this engine is currently being constructed and tested under the direction of the author at the Research Institute for Applied Knowledge Processing (FAW-Ulm) as Project ALIAS: Adaptive Learning Image Analysis System. Initial results indicate excellent detection, discrimination, and localization of anomalies.

A perspective on artificial intelligence: learning to learn

The classical approach to the acquisition of knowledge in artificial intelligence has been to program the intelligence into the machine in the form of specific rules for the application of the knowledge: expert systems. Unfortunately, the amount of time and resources required to program an expert system with sufficient knowledge for non-trivial problem-solving is prohibitively large. An alternative approach is to allow the machine tolearn the rules based upon trial-and-error interaction with the environment, much as humans do. This will require endowing the machine with a sophisticated set of sensors for the perception of the external world, the ability to generate trial actions based upon this perceived information, and a dynamic evaluation policy to allow it to measure the effectiveness of its trial actions and modify its repertoire accordingly. The principles underlying this paradigm, known ascollective learning systems theory, have already been applied to sophisticated gaming problems, demonstrating robust learning and dynamic adaptivity.The fundamental building block of a collective learning system is thelearning cell, which may be embedded in a massively parallel, hierarchical data communications network. Such a network comprising 100 million learning cells will approach the intelligence capacity of the human cortex. In the not-too-distant future, it may be possible to build a race of robotic slaves to perform a wide variety of tasks in our culture. This goal, while irresistibly attractive, is most certainly fraught with severe social, political, moral, and economic difficulties.

Speech Emotion Recognition using a backward context

The classification of emotions, such as joy, anger, anxiety, etc. from tonal variations in human speech is an important task for research and applications in human computer interaction. In the preceding work, it has been demonstrated that the locally extracted features of speech match or surpass the performance of global features that has been adopted in current approaches. In this continuing research, a backward context, which also can be considered as a feature vector memory, is shown to improve the prediction accuracy of the Speech Emotion Recognition engine. Preliminary results on German emotional speech database illustrate significant improvements over results from the previous study.

Using a Genetic Algorithm to Search for the Representational Bias of a Collective Reinforcement Learner

In reinforcement learning, the state generalization problem can be reformulated as the problem of finding a strong and correct representational bias for the learner. In this study, a genetic algorithm is used to find the representational bias of a stochastic reinforcement learner called a Collective Learning Automaton (CLA). The representational bias specifies an initial partition of the CLAs state space, which the CLA can strengthen during learning. The primary focus of this study is to investigate the usefulness of the very strong representational biases generated by the system. The study compares the accuracy of an inexperienced learners bias to the accuracy of an experienced learners bias using PAC- like measures of Valiant. The results presented in the paper demonstrate that, in general, it cannot be assumed that a representation that is part of an experienced learners solution to a problem will provide an accurate learning bias to an inexperienced learner.

Image-content classification using a dynamically allocated ALISA texture module

ALISA (adaptive learning image and signal analysis) is an adaptive learning image and signal classification engine based on the collective learning systems theory. Using supervised training, the ALISA engine builds a set of multidimensional feature histograms that estimate the joint probability density function of the feature space for each trained class. Six general-purpose features, one with a precision of 60 bins and the rest with 20 bins, were used to build a dynamically allocated sparse data structure instead of a complete static structure for each class. During the training of the new dynamically allocated ALISA with 6 different classes (sky, water, skin, rose, evergreen, and grass), a total about 12,000,000 counts were accumulated during training, generating fewer than 150,000 unique feature vectors. The results demonstrate the classification of several test images for each of the 6 trained classes. Much work remains to be done to optimize the new dynamically allocated ALISA classifier, but the initial results are encouraging.

A Performance Evaluation of ALIAS for the Detection of Geometric Anomalies on Fractal Images

Based on Collective Learning Systems Theory and a versatile general-purpose architecture for massively parallel networks of processors, a Transputer-based parallel-processing image-processing system, known as ALIAS (Adaptive Learning Image Analysis System), has been applied to a difficult image processing task: the phase, translation, and scale invariant detection of anomalous features in otherwise “normal” images. ALIAS consists of a three-layer hierarchical network of 32 learning cells and 33 non-learning cells. This paper presents an evaluation of ALIAS for the detection of anomalies which are square sections of a class of geometric images (called Manhattan) placed on binary fractal images belonging to one of three distinct “normal” equivalence classes (called Islands, Plateaus, and Wheatfields). Results indicate that the ability of ALIAS to detect and discriminate Manhattan anomalies improves with increasing area of the anomaly, although positive discrimination is only achieved with large anomalies. The ability of ALIAS to localize Manhattan anomalies exhibits a positive maximum when the area of the anomaly is 25 pixels, illustrating the Franz-Josef Syndrome.

A CLS hierarchy for the classification of images

The recognition of images beyond basic image processing often relies on training an adaptive system using a set of samples from a desired type of images. The adaptive algorithm used in this research is a learning automata model called CLS (collective learning systems). Using CLS, we propose a hierarchy of collective learning layers to learn color and texture feature patterns of images to perform three basic tasks: recognition, classification and segmentation. The higher levels in the hierarchy perform recognition, while the lower levels perform image segmentation. At the various levels the hierarchy is able to classify images according to learned patterns. In order to test the approach we use three examples of images: a) Satellite images of celestial planets, b) FFT spectral images of audio signals and c) family pictures for human skin recognition. By studying the multi-dimensional histogram of the selected images at each level we are able to determine the appropriate set of color and texture features to be used as input to a hierarchy of adaptive CLS to perform recognition and segmentation. Using the system in the proposed hierarchical manner, we obtained promising results that compare favorably with other AI approaches such as Neural Networks or Genetic Algorithms. “To understand is to perceive patterns” Sir Isaiah Berlin (1909-1997)