We present a new technique called “t-SNE” that visualizes high-dimensional data by giving each datapoint a location in a two or three-dimensional map. The technique is a variation of Stochastic Neighbor Embedding (Hinton and Roweis, 2002) that is much easier to optimize, and produces significantly better visualizations by reducing the tendency to crowd points together in the center of the map. t-SNE is better than existing techniques at creating a single map that reveals structure at many different scales. This is particularly important for high-dimensional data that lie on several different, but related, low-dimensional manifolds, such as images of objects from multiple classes seen from multiple viewpoints. For visualizing the structure of very large datasets, we show how t-SNE can use random walks on neighborhood graphs to allow the implicit structure of all of the data to influence the way in which a subset of the data is displayed. We illustrate the performance of t-SNE on a wide variety of datasets and compare it with many other non-parametric visualization techniques, including Sammon mapping, Isomap, and Locally Linear Embedding. The visualizations produced by t-SNE are significantly better than those produced by the other techniques on almost all of the datasets.

/pdf/visualizing-data-using-t-sne-2w1n304iq0.pdf

Visualizing Data using t-SNE

Three parallel algorithms for classical molecular dynamics are presented. The first assigns each processor a fixed subset of atoms; the second assigns each a fixed subset of inter-atomic forces to compute; the third assigns each a fixed spatial region. The algorithms are suitable for molecular dynamics models which can be difficult to parallelize efficiently—those with short-range forces where the neighbors of each atom change rapidly. They can be implemented on any distributed-memory parallel machine which allows for message-passing of data between independently executing processors. The algorithms are tested on a standard Lennard-Jones benchmark problem for system sizes ranging from 500 to 100,000,000 atoms on several parallel supercomputers--the nCUBE 2, Intel iPSC/860 and Paragon, and Cray T3D. Comparing the results to the fastest reported vectorized Cray Y-MP and C90 algorithm shows that the current generation of parallel machines is competitive with conventional vector supercomputers even for small problems. For large problems, the spatial algorithm achieves parallel efficiencies of 90% and a 1840-node Intel Paragon performs up to 165 faster than a single Cray C9O processor. Trade-offs between the three algorithms and guidelines for adapting them to more complex molecular dynamics simulations are also discussed.

Fast parallel algorithms for short-range molecular dynamics

https://axon.cs.byu.edu/~martinez/classes/678/Presentations/Yao.pdf

Extreme learning machine: Theory and applications

Probability Distributions.- Linear Models for Regression.- Linear Models for Classification.- Neural Networks.- Kernel Methods.- Sparse Kernel Machines.- Graphical Models.- Mixture Models and EM.- Approximate Inference.- Sampling Methods.- Continuous Latent Variables.- Sequential Data.- Combining Models.

Pattern Recognition and Machine Learning

Data Mining - Concepts and Techniques.

Une etape fondamentale dans la prediction d’une serie temporelle est la determination de son ordre auto-regressif et/ou d’une matrice auto-regressive. Nous proposons une methode de prediction basee sur la determination de la dimension fractale de la serie temporelle et sur une reduction de dimension par Analyse en Composantes Curvilignes (CCA). Les performances de la methode sont illustrees par deux exemples : une serie generee artificiellement et l’indice SBF 250.

/pdf/estimation-de-la-dimension-intrinseque-d-une-serie-3irm9dx6hu.pdf

Estimation de la dimension intrinsèque d'une série temporelle et prédiction par une méthode de projection. Application au SBF 250 sur la période 1992-1997

A business plan is a document presenting in a concise form the key elements (management, finance, marketing, ...) describing a percieved business opportunity. It is used among others as a tool for evaluating the feasibility and profitability of a project from the entrepreneur’ or investor’s point of view. In this work, we try to analyze to which degree a wellthought and well-written business plan leads to a successful venture in order to understand the link between business plan quality and venture success. We used a sample of 119 business plans that were collected in the context of a business plan competition organized in France, Belgium, Germany and Luxembourg. Each business plan was evaluated by experts, on a double-blind basis, according to seven criteria, with grades ranking from 0 to 10. Concerning the expected result of the classification, a project is considered successful if (1) it leads to an actual commercial activity and (2) it is still in activity 24 months after the initial assessment. When one tries to predict the success or failure of a project, one has to (1) find the most relevant criteria among the initial ones, (2) choose and design a classifier, and (3) choose a meaningfull threshold (cutoff value). The aim of the latest is to optimize an application-driven criterion, which might not be the classical ’least misclassification number’ or ’least mean square error’, while the first stage (variable selection) is necessary because of the small sample size, which forces to restrein the number of parameters in the models. To select the most useful variables (criteria), two approaches are commonly used. The first one consists in measuring the relevance of each single variable by evaluating its correlation (or in a more general manner, its mutual information) with the expected result of the classification. The second is based on the analysis of the performances of actual classifiers (in our case, a logistic regression and a k-NN model) built on each single variable. Note that both approaches can be upgraded to a constructive (iterative) method , where individual relevance is used at each step to add or subtract a variable. In order to compare non-linear classification methods to those classically used in entrepreneurship, we use two different models (techniques), namely a logistic regression (logit) and a k-nearest neighbour (k-NN). A resampling technique (leave-one-out) is used to validate them. Since not investing in a successfull venture is not as annoying as investing in a project that fails, the standard approach of minimizing the total number of bad classifications is certainly not a good criterion. Furthermore, the number of positive samples (success) being far lower than the number of negative samples (failures), minimizing misclassifications will eventually lead to a null-investment policy. As a consequence, we try to maximize criteria that are more application-specific, such as the precision or the recall. This is achieved by using an appropriate cutoff value (threshold) between the two classes at the output of the classifiers. In [1], the choice of a cutoff value was made according to the priors of each classes ( psuccess = 0.17), and a logit was used. We used a criterion-based cutoff choice for both a generalised linear model (logit) and a non linear one (k-NN). When precision is maximized with both models, the k-NN tends to reach a better recall than the logit, but its absolute precision is lower. The definitive choice between those models might be made using more domain-specific information such as the expected return-on-investment for each project.

Should Seed Investors Read Business Plans

An incremental version of the ELMVIS+ method is proposed in this paper. It iteratively selects a few best fitting data samples from a large pool, and adds them to the model. The method keeps high speed of ELMVIS+ while allowing for much larger possible sample pools due to lower memory requirements. The extension is useful for reaching a better local optimum with greedy optimization of ELMVIS, and the data structure can be specified in semi-supervised optimization. The major new application of incremental ELMVIS is not to visualization, but to a general dataset processing. The method is capable of learning dependencies from non-organized unsupervised data -- either reconstructing a shuffled dataset, or learning dependencies in complex high-dimensional space. The results are interesting and promising, although there is space for improvements.

Incremental ELMVIS for unsupervised learning

A plethora of methods have been developed to handle anomaly detection in various application domains. This work focuses on locating anomalies inside a categorical data set without assuming any spe- cific domain knowledge. By exploiting the conditional dependence and independence relationships among data attributes, not only can data an- alysts recognize the anomaly, but also locate the potentially anomalous attributes inside an anomalous instance following its masks. Masks are geometrically generated based on the factorization of the joint probability from a Bayesian network automatically learnt from the given data set.

Locating Anomalies Using Bayesian Factorizations and Masks

Multibeam bathymetry data could represent nearly continuous coverage depth measurements of the seafloor and reveal geomorphological regions. Recent studies have utilized multibeam bathymetry data to provide geological maps, but their delineations were done manually. Manual classification and delineation are inherently subjective and therefore can be inaccurate. In this paper, we try to develop one strategy to explore seafloor stretching in Mariana trench arc via squeeze and excitation network, combining data clustering, slope and gradient. In our experiments, we use the high-resolution multibeam bathymetric data collected by NOAA Office of Ocean Exploration and Research (OER). The geomorphological seabed in the Mariana region is automatically classified into different classes. The experimental results demonstrate that geomorphological seabed classification strategy achieves a robust, automated delineation approach.

Amaury Lendasse

Papers

Estimation de la dimension intrinsèque d'une série temporelle et prédiction par une méthode de projection. Application au SBF 250 sur la période 1992-1997

Should Seed Investors Read Business Plans

Incremental ELMVIS for unsupervised learning

Locating Anomalies Using Bayesian Factorizations and Masks

Exploring Seafloor Stretching in Mariana Trench Arc via the Squeeze and Excitation network with High-resolution Multibeam Bathymetric Survey