Jun Kitazono

Object Representation in Inferior Temporal Cortex Is Organized Hierarchically in a Mosaic-Like Structure

There are two dominant models for the functional organization of brain regions underlying object recognition. One model postulates category-specific modules while the other proposes a distributed representation of objects with generic visual features. Functional imaging techniques relying on metabolic signals, such as fMRI and optical intrinsic signal imaging (OISI), have been used to support both models, but due to the indirect nature of the measurements in these techniques, the existing data for one model cannot be used to support the other model. Here, we used large-scale multielectrode recordings over a large surface of anterior inferior temporal (IT) cortex, and densely mapped stimulus-evoked neuronal responses. We found that IT cortex is subdivided into distinct domains characterized by similar patterns of responses to the objects in our stimulus set. Each domain spanned several millimeters on the cortex. Some of these domains represented faces (“face” domains) or monkey bodies (“monkey-body” domains). We also identified domains with low responsiveness to faces (“anti-face” domains). Meanwhile, the recording sites within domains that displayed category selectivity showed heterogeneous tuning profiles to different exemplars within each category. This local heterogeneity was consistent with the stimulus-evoked feature columns revealed by OISI. Taken together, our study revealed that regions with common functional properties (domains) consist of a finer functional structure (columns) in anterior IT cortex. The “domains” and previously proposed “patches” are rather like “mosaics” where a whole mosaic is characterized by overall similarity in stimulus responses and pieces of the mosaic correspond to feature columns.

Novel method to classify hemodynamic response obtained using multi-channel fNIRS measurements into two groups: exploring the combinations of channels.

Near-infrared spectroscopy (NIRS) in psychiatric studies has widely demonstrated that cerebral hemodynamics differs among psychiatric patients. Recently we found that children with attention-deficit/hyperactivity disorder (ADHD) and children with autism spectrum disorders (ASD) showed different hemodynamic responses to their own mother’s face. Based on this finding, we may be able to classify the hemodynamic data into two those groups and predict to which diagnostic group an unknown participant belongs. In the present study, we proposed a novel statistical method for classifying the hemodynamic data of these two groups. By applying a support vector machine (SVM), we searched the combination of measurement channels at which the hemodynamic response differed between the ADHD and the ASD children. The SVM found the optimal subset of channels in each data set and successfully classified the ADHD data from the ASD data. For the 24-dimensional hemodynamic data, two optimal subsets classified the hemodynamic data with 84% classification accuracy, while the subset contained all 24 channels classified with 62% classification accuracy. These results indicate the potential application of our novel method for classifying the hemodynamic data into two groups and revealing the combinations of channels that efficiently differentiate the two groups.

Detection of DDoS Backscatter Based on Traffic Features of Darknet TCP Packets

In this work, we propose a method to discriminate backscatter caused by DDoS attacks from normal traffic. Since DDoS attacks are imminent threats which could give serious economic damages to private companies and public organizations, it is quite important to detect DDoS backscatter as early as possible. To do this, 11 features of port/IP information are defined for network packets which are sent within a short time, and these features of packet traffic are classified by Suppurt Vector Machine (SVM). In the experiments, we use TCP packets for the evaluation because they include control flags (e.g. SYN-ACK, RST-ACK, RST, ACK) which can give label information (i.e. Backscatter or non-backscatter). We confirm that the proposed method can discriminate DDoS backscatter correctly from unknown dark net TCP packets with more than 90% accuracy.

Efficient Algorithms for Searching the Minimum Information Partition in Integrated Information Theory

The ability to integrate information in the brain is considered to be an essential property for cognition and consciousness. Integrated Information Theory (IIT) hypothesizes that the amount of integrated information (Φ) in the brain is related to the level of consciousness. IIT proposes that, to quantify information integration in a system as a whole, integrated information should be measured across the partition of the system at which information loss caused by partitioning is minimized, called the Minimum Information Partition (MIP). The computational cost for exhaustively searching for the MIP grows exponentially with system size, making it difficult to apply IIT to real neural data. It has been previously shown that, if a measure of Φ satisfies a mathematical property, submodularity, the MIP can be found in a polynomial order by an optimization algorithm. However, although the first version of Φ is submodular, the later versions are not. In this study, we empirically explore to what extent the algorithm can be applied to the non-submodular measures of Φ by evaluating the accuracy of the algorithm in simulated data and real neural data. We find that the algorithm identifies the MIP in a nearly perfect manner even for the non-submodular measures. Our results show that the algorithm allows us to measure Φ in large systems within a practical amount of time.

Stochastic collapsed variational Bayesian inference for biterm topic model.

It is useful for many applications to find out meaningful topics from short texts, such as tweets and comments on websites. Since directly applying conventional topic models (e.g., LDA) to short texts often produces poor results, as a general approach to short texts, a biterm topic model (BTM) was recently proposed. However, the original BTM implementation uses collapsed Gibbs sampling (CGS) for its inference, which requires many iterations over the entire dataset. On the other hand, for LDA, there have been proposed many fast inference algorithms throughout the decade. Among them, a recently proposed stochastic collapsed variational Bayesian inference (SCVB0) is promising because it is applicable to an online setting and takes advantage of the collapsed representation, which results in an improved variational bound. Applying the idea of SCVB0, we develop a fast one-pass inference algorithm for BTM, which can be used to analyze large-scale general short texts and is extensible to an online setting. To evaluate the performance of the proposed algorithm, we conducted several experiments using short texts on Twitter. Experimental results showed that our algorithm found out meaningful topics significantly faster than the original algorithm.

Neural Network Model with Discrete and Continuous Information Representation

An associative memory model and a neural network model with a Mexican-hat type interaction are two major attractor neural network models. The associative memory model has discretely distributed fixed-point attractors, and achieves a discrete information representation. On the other hand, a neural network model with a Mexican-hat type interaction uses a ring attractor to achieves a continuous information representation, which can be seen in the working memory in the prefrontal cortex and columnar activity in the visual cortex. In the present study, we propose a neural network model that achieves discrete and continuous information representation. We use a statistical–mechanical analysis to find that a localized retrieval phase exists in the proposed model, where the memory pattern is retrieved in the localized subpopulation of the network. In the localized retrieval phase, the discrete and continuous information representation is achieved by using the orthogonality of the memory patterns and the neutral st...

Multidimensional Unfolding Based on Stochastic Neighbor Relationship

Multi-dimensional Unfolding (MU) is a method to visualize relevance data between two sets (e.g., preference data) as a single scatter plot. Usually, in the analysis of relevance data, users are interested in which elements are strongly related to each other (e.g., how much an individual likes an item), and not in which elements are irrelevant to each other. However, the conventional MU often suffers from the problem that relationships between irrelevant pairs are overly emphasized and those between relevant pairs are not represented appropriately. Here we propose novel MU methods based on stochastic neighbor relationship, by extending dimensionality reduction methods, Stochastic Neighbor Embed- ding (SNE) and t-distributed SNE. The proposed methods are defined by Kullback-Leibler divergence (KL divergence), and because of the asymmetric property of KL divergence, they give priority to representing relationships between relevant pairs. Experimental results show that the proposed methods can alleviate the problem and achieve reasonable visualization compared to the conventional MU.

t-Distributed Stochastic Neighbor Embedding with Inhomogeneous Degrees of Freedom

One of the dimension reduction (DR) methods for data-visualization, t-distributed stochastic neighbor embedding (t-SNE), has drawn increasing attention. t-SNE gives us better visualization than conventional DR methods, by relieving so-called crowding problem. The crowding problem is one of the curses of dimensionality, which is caused by discrepancy between high and low dimensional spaces. However, in t-SNE, it is assumed that the strength of the discrepancy is the same for all samples in all datasets regardless of ununiformity of distributions or the difference in dimensions, and this assumption sometimes ruins visualization. Here we propose a new DR method inhomogeneous t-SNE, in which the strength is estimated for each point and dataset. Experimental results show that such pointwise estimation is important for reasonable visualization and that the proposed method achieves better visualization than the original t-SNE.

A sentiment polarity prediction model using transfer learning and its application to SNS flaming event detection

In recent years, with the popularization of SNS, the incidents called flaming, in which a large number of negative comments are retweeted and spread to many followers on SNS, are increasing. Since a flaming event sometimes causes severe criticism by public people, it is becoming a great thread to companies and therefore it is important for companies to protect their reputation from such flaming events. In order to protect companies from serious damages in reputation, we propose a machine learning approach to the detection of flaming events by monitoring the sentiment polarity of SNS comments. From the nature of SNS comments such as the spread of a large number of retweets with the same content for a short time, the word distributions are often strongly biased and it leads to poor performance in sentiment polarity prediction. To alleviate this problem, we introduce transfer learning into the conventional Naive Bayes classifier. More concretely, in the Naive Bayes classifier, the occurrence probabilities of words on a target domain are recalculated using those on other domains, where a domain corresponds to a company to be protected. The experimental results demonstrate that the proposed transfer learning contribute to the improvement in the sentiment polarity prediction for SNS comments. In addition, we show that the proposed system can detect flaming events correctly by monitoring the number of negative comments.

t-Distributed stochastic neighbor embedding spectral clustering

This paper introduces a new topological clustering approach to cluster high dimensional datasets based on t-SNE (Stochastic Neighbor Embedding) dimensionality reduction method and spectral clustering. Spectral clustering method needs to construct an adjacency matrix and calculate the eigen-decomposition of the corresponding Laplacian matrix [1] which are computational expensive and is not easy to apply on large-scale data sets. One of the issue of this problem is to reduce the dimensionality befor to cluster the dataset. The t-SNE method which performs good results for visulaization allows a projection of the dataset in low dimensional spaces that make it easy to use for very large datasets. Using t-SNE during the learning process will allow to reduce the dimensionality and to preserve the topology of the dataset by increasing the clustering accuracy. We illustrate the power of this method with several real datasets. The results show a good quality of clustering results and a higher speed.