Neuroscientists are developing new imaging techniques and generating large volumes of data in an ... more Neuroscientists are developing new imaging techniques and generating large volumes of data in an effort to understand the complex structure of the nervous system. The complexity and size of this data makes human interpretation a labor-intensive task. To aid in the analysis, new segmentation techniques for identifying neurons in these feature rich datasets are required. This paper presents a method for neuron boundary detection and nonbranching process segmentation in electron microscopy images and visualizing them in three dimensions. It combines both automated segmentation techniques with a graphical user interface for correction of mistakes in the automated process. The automated process first uses machine learning and image processing techniques to identify neuron membranes that deliniate the cells in each two-dimensional section. To segment nonbranching processes, the cell regions in each two-dimensional section are connected in 3D using correlation of regions between sections. The combination of this method with a graphical user interface specially designed for this purpose, enables users to quickly segment cellular processes in large volumes.
This chapter contains sections titled: Introduction Historical Overview of BMI Approaches/Models ... more This chapter contains sections titled: Introduction Historical Overview of BMI Approaches/Models Characteristics of Neural Recordings Modeling Problem Examples Performance Discussion References
In this paper, we propose a graph-theoretical approach to reveal patterns of functional dependenc... more In this paper, we propose a graph-theoretical approach to reveal patterns of functional dependencies between different scalp regions. We start by computing pairwise measures of dependence from dense-array scalp electroencephalographic (EEG) recordings. The obtained dependence matrices are then averaged over trials and further statistically processed to provide more reliability. Graph structure information is subsequently extracted using several graph theoretical measures. Simple measures of node degree and clustering strength are shown to be useful to describe the global properties of the analyzed networks. More sophisticated measures, such as betweenness centrality and subgraph centrality tend to provide additional insight into the network structure, and therefore robustly discriminate two cognitive states. We further examine the connected components of the graph to identify the dependent functional regions. The approach supports dynamicity in that all suggested computations can be easily extended to different points in time, thus enabling to monitor dependence evolution and variability with time.
In this paper, we study the mean square convergence of the kernel least mean square (KLMS). The f... more In this paper, we study the mean square convergence of the kernel least mean square (KLMS). The fundamental energy conservation relation has been established in feature space. Starting from the energy conservation relation, we carry out the mean square convergence analysis and obtain several important theoretical results, including an upper bound on step size that guarantees the mean square convergence, the theoretical steady-state excess mean square error (EMSE), an optimal step size for the fastest convergence, and an optimal kernel size for the fastest initial convergence. Monte Carlo simulation results agree with the theoretical analysis very well.
IEEE transactions on neural networks and learning systems, Sep 1, 2013
In a recent paper, we developed a novel quantized kernel least mean square algorithm, in which th... more In a recent paper, we developed a novel quantized kernel least mean square algorithm, in which the input space is quantized (partitioned into smaller regions) and the network size is upper bounded by the quantization codebook size (number of the regions). In this brief, we propose the quantized kernel least squares regression, and derive the optimal solution. By incorporating a simple online vector quantization method, we derive a recursive algorithm to update the solution, namely the quantized kernel recursive least squares algorithm. The good performance of the new algorithm is demonstrated by Monte Carlo simulations.
Inspired by the human vision system and learning, we propose a novel cognitive architecture that ... more Inspired by the human vision system and learning, we propose a novel cognitive architecture that understands the content of raw videos in terms of objects without using labels. The architecture achieves four objectives: (1) Decomposing raw frames in objects by exploiting foveal vision and memory. (2) Describing the world by projecting objects on an internal canvas. (3) Extracting relevant objects from the canvas by analyzing the causal relation between objects and rewards. (4) Exploiting the information of relevant objects to facilitate the reinforcement learning (RL) process. In order to speed up learning, and better identify objects that produce rewards, the architecture implements learning by causality from the perspective of Wiener and Granger using object trajectories stored in working memory and the time series of external rewards. A novel non-parametric estimator of directed information using Renyi's entropy is designed and tested. Experiments on three environments show that our architecture extracts most of relevant objects. It can be thought of as 'understanding' the world in an object-oriented way. As a consequence, our architecture outperforms state-of-the-art deep reinforcement learning in terms of training speed and transfer learning.
Deep learning architectures are an extremely powerful tool for recognizing and classifying images... more Deep learning architectures are an extremely powerful tool for recognizing and classifying images. However, they require supervised learning and normally work on vectors of the size of image pixels and produce the best results when trained on millions of object images. To help mitigate these issues, we propose an end-to-end architecture that fuses bottom-up saliency and top-down attention with an object recognition module to focus on relevant data and learn important features that can later be fine-tuned for a specific task, employing only unsupervised learning. In addition, by utilizing a virtual fovea that focuses on relevant portions of the data, the training speed can be greatly improved. We test the performance of the proposed Gamma saliency technique on the Toronto and CAT 2000 databases, and the foveated vision in the large Street View House Numbers (SVHN) database. The results with foveated vision show that Gamma saliency performs at the same level as the best alternative algorithms while being computationally faster. The results in SVHN show that our unsupervised cognitive architecture is comparable to fully supervised methods and that saliency also improves CNN performance if desired. Finally, we develop and test a top-down attention mechanism based on the Gamma saliency applied to the top layer of CNNs to facilitate scene understanding in multi-object cluttered images. We show that the extra information from top-down saliency is capable of speeding up the extraction of digits in the cluttered multidigit MNIST data set, corroborating the important role of top down attention.
In modern science, engineering and public life, a big role is played by information and operation... more In modern science, engineering and public life, a big role is played by information and operations associated with it: information reception, information transmission, information processing, storing information and so on. The significance of information has seemingly outgrown the significance of the other important factor, which used to play a dominant role in the previous century, namely, energy.
This chapter is devoted to the second variational problem, in which we try to find an extremum of... more This chapter is devoted to the second variational problem, in which we try to find an extremum of the Shannon’s amount of information with respect to different input distributions. We assume that the channel, i.e. a conditional distribution on its output with a fixed input signal, is known. The maximum amount of information between the input and output signals is called channel capacity. Contrary to the conventional presentation, from the very beginning we introduce an additional constraint concerning the mean value of some function of input variables, i.e. we consider a conditional variational problem. Results for the case without the constraint are obtained as a particular case of the provided general results.
The definition of the amount of information, given in Chapter 1, is justified when we deal with a... more The definition of the amount of information, given in Chapter 1, is justified when we deal with a transformation of information from one kind into another, i.e. when considering encoding of information. It is essential that the law of conservation of information amount holds under such a transformation. It is very useful to draw an analogy with the law of conservation of energy. The latter is the main argument for introducing the notion of energy. Of course, the law of conservation of information is more complex than the law of conservation of energy in two respects. The law of conservation of energy establishes an exact equality of energies, when one type of energy is transformed into another. However, in transforming information we have a more complex relation, namely ‘not greater’ (\(\leqslant \)), i.e. the amount of information cannot increase. The equality sign corresponds to optimal encoding. Thus, when formulating the law of conservation of information, we have to point out that there possibly exists such an encoding, for which the equality of the amounts of information occurs.
In brain-machine interface (BMI) modeling, the firing patterns of hundreds of neurons are used to... more In brain-machine interface (BMI) modeling, the firing patterns of hundreds of neurons are used to reconstruct a variety of kinematic variables. The large number of neurons produces an explosion in the number of free parameters, which affects model generalization. This paper proposes a model-free measure of pairwise neural dependence to rank the importance of neurons in neural to motor mapping. Compared to a model-dependent approach such as sensitivity analysis, sixty percent of the neurons with the strongest dependence coincide with the top 10 most sensitive neurons trained through the model. Using this data-driven approach that operates on the input data alone, it is possible to perform neuron selection in a more efficient way that is not subject to assumptions about decoding models. To further understand the functional dependencies that influence neural to motor mapping, we use an open source available graph visualization toolkit called Prefuse to visualize the neural dependency graph and quantify the functional connectivity in motor cortex. This tool when adapted to the analysis of neuronal recordings has the potential to easily display the relationships in data of large dimension.
CiteSeer X (The Pennsylvania State University), 2012
The purpose of this paper is twofold: first, to propose a modification to the generalized measure... more The purpose of this paper is twofold: first, to propose a modification to the generalized measure of association (GMA) framework that reduces the effect of temporal structure in time series; second, to assess the reliability of using association methods to capture dependence between pairs of EEG channels using their time series or envelopes. To achieve the first goal, the GMA algorithm was updated so as to minimize the effect of the correlation inherent in the time structure. The reliability of the modified scheme was then assessed on both synthetic and real data. Synthetic data was generated from a Clayton copula, for which null hypotheses of uncorrelatedness were constructed for the signal. The signal was processed such that the envelope emulated important characteristics of experimental EEG data. Results show that the modified GMA procedure can capture pairwise dependence between generated signals as well as their envelopes with good statistical power. Furthermore, applying GMA and Kendall's tau to quantify dependence using the extracted envelopes of processed EEG data concords with previous findings using the signal itself.
In this paper, a novel extended kernel recursive least squares algorithm is proposed combining th... more In this paper, a novel extended kernel recursive least squares algorithm is proposed combining the kernel recursive least squares algorithm and the Kalman filter or its extensions to estimate or predict signals. Unlike the extended kernel recursive least squares (Ex-KRLS) algorithm proposed by Liu, the state model of our algorithm is still constructed in the original state space and the hidden state is estimated using the Kalman filter. The measurement model used in hidden state estimation is learned by the kernel recursive least squares algorithm (KRLS) in reproducing kernel Hilbert space (RKHS). The novel algorithm has more flexible state and noise models. We apply this algorithm to vehicle tracking and the nonlinear Rayleigh fading channel tracking, and compare the tracking performances with other existing algorithms.
ABSTRACT Similarity measures play a critical role in the solution quality of data analysis method... more ABSTRACT Similarity measures play a critical role in the solution quality of data analysis methods. Outliers or noise often taint the solution, hence, practical data analysis calls for robust measures. The correntropic loss function is a smooth and robust measure. In this paper, we present the properties of the correntropic loss function that can be utilized in optimization based data analysis methods.
Neuroscientists are developing new imaging techniques and generating large volumes of data in an ... more Neuroscientists are developing new imaging techniques and generating large volumes of data in an effort to understand the complex structure of the nervous system. The complexity and size of this data makes human interpretation a labor-intensive task. To aid in the analysis, new segmentation techniques for identifying neurons in these feature rich datasets are required. This paper presents a method for neuron boundary detection and nonbranching process segmentation in electron microscopy images and visualizing them in three dimensions. It combines both automated segmentation techniques with a graphical user interface for correction of mistakes in the automated process. The automated process first uses machine learning and image processing techniques to identify neuron membranes that deliniate the cells in each two-dimensional section. To segment nonbranching processes, the cell regions in each two-dimensional section are connected in 3D using correlation of regions between sections. The combination of this method with a graphical user interface specially designed for this purpose, enables users to quickly segment cellular processes in large volumes.
This chapter contains sections titled: Introduction Historical Overview of BMI Approaches/Models ... more This chapter contains sections titled: Introduction Historical Overview of BMI Approaches/Models Characteristics of Neural Recordings Modeling Problem Examples Performance Discussion References
In this paper, we propose a graph-theoretical approach to reveal patterns of functional dependenc... more In this paper, we propose a graph-theoretical approach to reveal patterns of functional dependencies between different scalp regions. We start by computing pairwise measures of dependence from dense-array scalp electroencephalographic (EEG) recordings. The obtained dependence matrices are then averaged over trials and further statistically processed to provide more reliability. Graph structure information is subsequently extracted using several graph theoretical measures. Simple measures of node degree and clustering strength are shown to be useful to describe the global properties of the analyzed networks. More sophisticated measures, such as betweenness centrality and subgraph centrality tend to provide additional insight into the network structure, and therefore robustly discriminate two cognitive states. We further examine the connected components of the graph to identify the dependent functional regions. The approach supports dynamicity in that all suggested computations can be easily extended to different points in time, thus enabling to monitor dependence evolution and variability with time.
In this paper, we study the mean square convergence of the kernel least mean square (KLMS). The f... more In this paper, we study the mean square convergence of the kernel least mean square (KLMS). The fundamental energy conservation relation has been established in feature space. Starting from the energy conservation relation, we carry out the mean square convergence analysis and obtain several important theoretical results, including an upper bound on step size that guarantees the mean square convergence, the theoretical steady-state excess mean square error (EMSE), an optimal step size for the fastest convergence, and an optimal kernel size for the fastest initial convergence. Monte Carlo simulation results agree with the theoretical analysis very well.
IEEE transactions on neural networks and learning systems, Sep 1, 2013
In a recent paper, we developed a novel quantized kernel least mean square algorithm, in which th... more In a recent paper, we developed a novel quantized kernel least mean square algorithm, in which the input space is quantized (partitioned into smaller regions) and the network size is upper bounded by the quantization codebook size (number of the regions). In this brief, we propose the quantized kernel least squares regression, and derive the optimal solution. By incorporating a simple online vector quantization method, we derive a recursive algorithm to update the solution, namely the quantized kernel recursive least squares algorithm. The good performance of the new algorithm is demonstrated by Monte Carlo simulations.
Inspired by the human vision system and learning, we propose a novel cognitive architecture that ... more Inspired by the human vision system and learning, we propose a novel cognitive architecture that understands the content of raw videos in terms of objects without using labels. The architecture achieves four objectives: (1) Decomposing raw frames in objects by exploiting foveal vision and memory. (2) Describing the world by projecting objects on an internal canvas. (3) Extracting relevant objects from the canvas by analyzing the causal relation between objects and rewards. (4) Exploiting the information of relevant objects to facilitate the reinforcement learning (RL) process. In order to speed up learning, and better identify objects that produce rewards, the architecture implements learning by causality from the perspective of Wiener and Granger using object trajectories stored in working memory and the time series of external rewards. A novel non-parametric estimator of directed information using Renyi's entropy is designed and tested. Experiments on three environments show that our architecture extracts most of relevant objects. It can be thought of as 'understanding' the world in an object-oriented way. As a consequence, our architecture outperforms state-of-the-art deep reinforcement learning in terms of training speed and transfer learning.
Deep learning architectures are an extremely powerful tool for recognizing and classifying images... more Deep learning architectures are an extremely powerful tool for recognizing and classifying images. However, they require supervised learning and normally work on vectors of the size of image pixels and produce the best results when trained on millions of object images. To help mitigate these issues, we propose an end-to-end architecture that fuses bottom-up saliency and top-down attention with an object recognition module to focus on relevant data and learn important features that can later be fine-tuned for a specific task, employing only unsupervised learning. In addition, by utilizing a virtual fovea that focuses on relevant portions of the data, the training speed can be greatly improved. We test the performance of the proposed Gamma saliency technique on the Toronto and CAT 2000 databases, and the foveated vision in the large Street View House Numbers (SVHN) database. The results with foveated vision show that Gamma saliency performs at the same level as the best alternative algorithms while being computationally faster. The results in SVHN show that our unsupervised cognitive architecture is comparable to fully supervised methods and that saliency also improves CNN performance if desired. Finally, we develop and test a top-down attention mechanism based on the Gamma saliency applied to the top layer of CNNs to facilitate scene understanding in multi-object cluttered images. We show that the extra information from top-down saliency is capable of speeding up the extraction of digits in the cluttered multidigit MNIST data set, corroborating the important role of top down attention.
In modern science, engineering and public life, a big role is played by information and operation... more In modern science, engineering and public life, a big role is played by information and operations associated with it: information reception, information transmission, information processing, storing information and so on. The significance of information has seemingly outgrown the significance of the other important factor, which used to play a dominant role in the previous century, namely, energy.
This chapter is devoted to the second variational problem, in which we try to find an extremum of... more This chapter is devoted to the second variational problem, in which we try to find an extremum of the Shannon’s amount of information with respect to different input distributions. We assume that the channel, i.e. a conditional distribution on its output with a fixed input signal, is known. The maximum amount of information between the input and output signals is called channel capacity. Contrary to the conventional presentation, from the very beginning we introduce an additional constraint concerning the mean value of some function of input variables, i.e. we consider a conditional variational problem. Results for the case without the constraint are obtained as a particular case of the provided general results.
The definition of the amount of information, given in Chapter 1, is justified when we deal with a... more The definition of the amount of information, given in Chapter 1, is justified when we deal with a transformation of information from one kind into another, i.e. when considering encoding of information. It is essential that the law of conservation of information amount holds under such a transformation. It is very useful to draw an analogy with the law of conservation of energy. The latter is the main argument for introducing the notion of energy. Of course, the law of conservation of information is more complex than the law of conservation of energy in two respects. The law of conservation of energy establishes an exact equality of energies, when one type of energy is transformed into another. However, in transforming information we have a more complex relation, namely ‘not greater’ (\(\leqslant \)), i.e. the amount of information cannot increase. The equality sign corresponds to optimal encoding. Thus, when formulating the law of conservation of information, we have to point out that there possibly exists such an encoding, for which the equality of the amounts of information occurs.
In brain-machine interface (BMI) modeling, the firing patterns of hundreds of neurons are used to... more In brain-machine interface (BMI) modeling, the firing patterns of hundreds of neurons are used to reconstruct a variety of kinematic variables. The large number of neurons produces an explosion in the number of free parameters, which affects model generalization. This paper proposes a model-free measure of pairwise neural dependence to rank the importance of neurons in neural to motor mapping. Compared to a model-dependent approach such as sensitivity analysis, sixty percent of the neurons with the strongest dependence coincide with the top 10 most sensitive neurons trained through the model. Using this data-driven approach that operates on the input data alone, it is possible to perform neuron selection in a more efficient way that is not subject to assumptions about decoding models. To further understand the functional dependencies that influence neural to motor mapping, we use an open source available graph visualization toolkit called Prefuse to visualize the neural dependency graph and quantify the functional connectivity in motor cortex. This tool when adapted to the analysis of neuronal recordings has the potential to easily display the relationships in data of large dimension.
CiteSeer X (The Pennsylvania State University), 2012
The purpose of this paper is twofold: first, to propose a modification to the generalized measure... more The purpose of this paper is twofold: first, to propose a modification to the generalized measure of association (GMA) framework that reduces the effect of temporal structure in time series; second, to assess the reliability of using association methods to capture dependence between pairs of EEG channels using their time series or envelopes. To achieve the first goal, the GMA algorithm was updated so as to minimize the effect of the correlation inherent in the time structure. The reliability of the modified scheme was then assessed on both synthetic and real data. Synthetic data was generated from a Clayton copula, for which null hypotheses of uncorrelatedness were constructed for the signal. The signal was processed such that the envelope emulated important characteristics of experimental EEG data. Results show that the modified GMA procedure can capture pairwise dependence between generated signals as well as their envelopes with good statistical power. Furthermore, applying GMA and Kendall's tau to quantify dependence using the extracted envelopes of processed EEG data concords with previous findings using the signal itself.
In this paper, a novel extended kernel recursive least squares algorithm is proposed combining th... more In this paper, a novel extended kernel recursive least squares algorithm is proposed combining the kernel recursive least squares algorithm and the Kalman filter or its extensions to estimate or predict signals. Unlike the extended kernel recursive least squares (Ex-KRLS) algorithm proposed by Liu, the state model of our algorithm is still constructed in the original state space and the hidden state is estimated using the Kalman filter. The measurement model used in hidden state estimation is learned by the kernel recursive least squares algorithm (KRLS) in reproducing kernel Hilbert space (RKHS). The novel algorithm has more flexible state and noise models. We apply this algorithm to vehicle tracking and the nonlinear Rayleigh fading channel tracking, and compare the tracking performances with other existing algorithms.
ABSTRACT Similarity measures play a critical role in the solution quality of data analysis method... more ABSTRACT Similarity measures play a critical role in the solution quality of data analysis methods. Outliers or noise often taint the solution, hence, practical data analysis calls for robust measures. The correntropic loss function is a smooth and robust measure. In this paper, we present the properties of the correntropic loss function that can be utilized in optimization based data analysis methods.
Uploads
Papers by Jose Principe