Emre Aksan

Learning Deep Temporal Representations for fMRI Brain Decoding

Functional magnetic resonance imaging fMRI produces low number of samples in high dimensional vector spaces which is hardly adequate for brain decoding tasks. In this study, we propose a combination of autoencoding and temporal convolutional neural network architecture which aims to reduce the feature dimensionality along with improved classification performance. The proposed network learns temporal representations of voxel intensities at each layer of the network by leveraging unlabeled fMRI data with regularized autoencoders. Learned temporal representations capture the temporal regularities of the fMRI data and are observed to be an expressive bank of activation patterns. Then a temporal convolutional neural network with spatial pooling layers reduces the dimensionality of the learned representations. By employing the proposed method, raw input fMRI data is mapped to a low-dimensional feature space where the final classification is conducted. In addition, a simple decorrelated representation approach is proposed for tuning the model hyper-parameters. The proposed method is tested on a ten class recognition memory experiment with nine subjects. Results support the efficiency and potential of the proposed model, compared to the baseline multi-voxel pattern analysis techniques.

LARGE SCALE FUNCTIONAL CONNECTIVITY FOR BRAIN DECODING

Functional Magnetic Resonance Imaging (fMRI) data consists of time series for each voxel recorded during a cognitive task. In order to extract useful information from this noisy and redundant data, techniques are proposed to select the voxels that are relevant to the underlying cognitive task. We propose a simple and efficient algorithm for decoding the brain states by modelling the correlation patterns between the voxel time series. For each stimulus during the experiment, a separate functional connectivity matrix is computed in voxel level. The elements in connectivity matrices are then filtered out by making use of a minimum spanning tree formed using a global connectivity matrix for the entire experiment in order to reduce dimensionality. For a recognition memory experiment with nine subjects, functional connectivity matrices are computed for encoding and retrieval phases. The class labels of the retrieval samples are predicted within a k-nearest neighbour space constructed by the traversed entries in the functional connectivity matrices for encoding samples. The proposed method is also adapted to large scale functional connectivity tasks by making use of graphics boards. Classification performance in ten categories is comparable and even better compared to both classical and enhanced methods of multi-voxel pattern analysis techniques.

Functional networks of anatomic brain regions

In this study, we propose a new approach to construct a two-level functional brain network. The nodes of the first-level network are the voxels of the functional Magnetic Resonance Images (fMRI) recorded during an object recognition task. The nodes of the network at the second-level are the anatomic regions of the brain. The arcs of the first level are estimated by a linear regression equation for the meshes formed around each voxel. Neighbors of each voxel are determined by using a functional similarity metric. The node degree distributions of the voxel-level functional brain network are then used to estimate the node attributes and arc weights between the nodes of anatomic regions at the second level. The region-level functional brain network is then used to analyze the relationship among the anatomic regions of the brain during a cognitive process. Our results indicate that, although the neighborhood is defined functionally, voxels tend to make connections within the anatomic regions. Therefore, it can be deduced that nearby voxels work coherently during the cognitive task compared to the voxels apart from each other.

DeepWriting: Making Digital Ink Editable via Deep Generative Modeling

Digital ink promises to combine the flexibility and aesthetics of handwriting and the ability to process, search and edit digital text. Character recognition converts handwritten text into a digital representation, albeit at the cost of losing personalized appearance due to the technical difficulties of separating the interwoven components of content and style. In this paper, we propose a novel generative neural network architecture that is capable of disentangling style from content and thus making digital ink editable. Our model can synthesize arbitrary text, while giving users control over the visual appearance (style). For example, allowing for style transfer without changing the content, editing of digital ink at the word level and other application scenarios such as spell-checking and correction of handwritten text. We furthermore contribute a new dataset of handwritten text with fine-grained annotations at the character level and report results from an initial user evaluation.

Guiding InfoGAN with Semi-supervision

In this paper we propose a new semi-supervised GAN architecture (ss-InfoGAN) for image synthesis that leverages information from few labels (as little as 0.22%, max. 10% of the dataset) to learn semantically meaningful and controllable data representations where latent variables correspond to label categories. The architecture builds on Information Maximizing Generative Adversarial Networks (InfoGAN) and is shown to learn both continuous and categorical codes and achieves higher quality of synthetic samples compared to fully unsupervised settings. Furthermore, we show that using small amounts of labeled data speeds-up training convergence. The architecture maintains the ability to disentangle latent variables for which no labels are available. Finally, we contribute an information-theoretic reasoning on how introducing semi-supervision increases mutual information between synthetic and real data.

Bilişsel durum analizi i~in beyin Aği modeli

We suggest a new approach to estimate a brain network to model cognitive tasks and explore the node degree distribution of this network in anatomic regions. Functional Magnetic Resonance Images are used to estimate the relationship among the voxels. First, a local mesh is formed around each voxel in a predefined neighborhood system. Then, the edge weights of meshes, called Mesh Arc Descriptors (MAD) are estimated using a linear regression model. In order to estimate the optimal mesh size for voxels, the error term obtained during the estimation of Mesh Arc Descriptors are employed to optimize Akaikes Information Criterion. Finally, the brain network is constructed for each class by the estimated MAD. During experiments, we analyze how the degree of nodes varies across the anatomic brain regions for different cognitive states. Our results indicate that some anatomic regions make dense connections for all cognitive tasks whereas some of them have relatively sparse connections. This observation is consistent with the previously reported findings of anatomic regions. Although the degree distributions look similar for all classes, there are slight variations among classes. Therefore, the statistics of node degree distribution may be used to discriminate the anatomic regions related to cognitive tasks.

Estimating brain connectivity for pattern analysis

In this study, the degree of connectivity for each voxel, which is the unit element of functional Magnetic Resonance Imaging (fMRI) data, with its neighboring voxels is estimated. The neighborhood system is defined by spatial connectivity metrics and a local mesh of variable size is formed around each voxel using spatial neighborhood. Then, the mesh arc weights, called Mesh Arc Descriptors (MAD), are used to represent each voxel rather than its own intensity value measured by functional Magnetic Resonance Images (fMRI). Finally, the optimal mesh size of each voxel is estimated using various information theoretic criteria. fMRI measurements are obtained during a memory encoding and retrieval experiment performed on a subject who is exposed to the stimuli from 10 semantic categories. Using the Mesh Arc Descriptors (MAD) having the variable mesh sizes, a k-NN classifier is trained. The classification performances reflect that the suggested variable-size Mesh Arc Descriptors represent the cognitive states better than the classical multi-voxel pattern representation and fixed-size Mesh Arc Descriptors. Moreover, it is observed that the degree of connectivities in the brain greatly varies for each voxel.

Modeling the Brain Connectivity for Pattern Analysis

An information theoretic approach is proposed to estimate the degree of connectivity for each voxel with its neighboring voxels. The neighborhood system is defined by spatial and functional connectivity metrics. Then, a local mesh of variable size is formed around each voxel using spatial or functional neighborhood. The mesh arc weights, called Mesh Arc Descriptors (MAD), are estimated by a linear regression model fitted to the voxel intensity values of the functional Magnetic Resonance Images (fMRI). Finally, the error term of the linear regression equation is used to estimate the mesh size for a voxel by optimizing Akaikes information Criterion, Bayesian Information Criterion and Rissanens Minimum Description Length. fMRI measurements are obtained during a memory encoding and retrieval experiment performed on a subject who is exposed to the stimuli from 10 semantic categories. For each sample, a k-NN classifier is trained using the Mesh Arc Descriptors (MAD) having the variable mesh sizes. The classification performances reflect that the suggested variable-size Mesh Arc Descriptors represents the mental states better than the classical multi-voxel pattern representation. Moreover, we observe that the degree of connectivities in the brain greatly varies for each voxel.