Abstract: t-SNE is an embedding method that the data science community has widely used. It helps two main tasks: to display results by coloring items according to the item class or feature value; and for forensic, giving a first overview of the dataset distribution. Two interesting characteristics of t-SNE are the structure preservation property and the answer to the crowding problem, where all neighbors in high dimensional space cannot be represented correctly in low dimensional space. t-SNE preserves the local neighborhood, and similar items are nicely spaced by adjusting to the local density. These two characteristics produce a meaningful representation, where the cluster area is proportional to its size in number, and relationships between clusters are materialized by closeness on the embedding. This algorithm is non-parametric. The transformation from a high to low dimensional space is described but not learned. Two initializations of the algorithm would lead to two different embedding. In a forensic approach, analysts would like to compare two or more datasets using their embedding. A naive approach would be to embed all datasets together. However, this process is costly as the complexity of t-SNE is quadratic, and would be infeasible for too many datasets. Another approach would be to learn a parametric model over an embedding built with a subset of data. While this approach is highly scalable, points could be mapped at the same exact position, making them indistinguishable. This type of model would be unable to adapt to new outliers nor concept drift. This paper presents a methodology to reuse an embedding to create a new one, where cluster positions are preserved. The optimization process minimizes two costs, one relative to the embedding shape and the second relative to the support embedding’ match. The embedding with the support process can be repeated more than once, with the newly obtained embedding. The successive embedding can be used to study the impact of one variable over the dataset distribution or monitor changes over time. This method has the same complexity as t-SNE per embedding, and memory requirements are only doubled. For a dataset of n elements sorted and split into k subsets, the total embedding complexity would be reduced from O(n2) to O(n2/k), and the memory requirement from n2 to 2(n/k)2 which enables computation on recent laptops. The method showed promising results on a real-world dataset, allowing to observe the birth, evolution and death of clusters. The proposed approach facilitates identifying significant trends and changes, which empowers the monitoring high dimensional datasets’ dynamics.
Abstract: Clustering is an unsupervised learning technique for aggregating data objects into meaningful classes so that intra cluster similarity is maximized and inter cluster similarity is minimized in data mining. However, no single clustering algorithm proves to be the most effective in producing the best result. As a result, a new challenging technique known as the cluster ensemble approach has blossomed in order to determine the solution to this problem. For the cluster analysis issue, this new technique is a successful approach. The cluster ensemble's main goal is to combine similar clustering solutions in a way that achieves the precision while also improving the quality of individual data clustering. Because of the massive and rapid creation of new approaches in the field of data mining, the ongoing interest in inventing novel algorithms necessitates a thorough examination of current techniques and future innovation. This paper presents a comparative analysis of various cluster ensemble approaches, including their methodologies, formal working process, and standard accuracy and error rates. As a result, the society of clustering practitioners will benefit from this exploratory and clear research, which will aid in determining the most appropriate solution to the problem at hand.
Abstract: Intelligent systems are required in order to quickly and accurately analyze enormous quantities of data in the Internet environment. In intelligent systems, information extracting processes can be divided into supervised learning and unsupervised learning. This paper investigates intelligent clustering by unsupervised learning. Intelligent clustering is the clustering system which determines the clustering model for data analysis and evaluates results by itself. This system can make a clustering model more rapidly, objectively and accurately than an analyzer. The methodology for the automatic clustering intelligent system is a multi-agent system that comprises a clustering agent and a cluster performance evaluation agent. An agent exchanges information about clusters with another agent and the system determines the optimal cluster number through this information. Experiments using data sets in the UCI Machine Repository are performed in order to prove the validity of the system.
Abstract: In this paper, a model of self-organizing spiking neural networks is introduced and applied to mobile robot environment representation and path planning problem. A network of spike-response-model neurons with a recurrent architecture is used to create robot-s internal representation from surrounding environment. The overall activity of network simulates a self-organizing system with unsupervised learning. A modified A* algorithm is used to find the best path using this internal representation between starting and goal points. This method can be used with good performance for both known and unknown environments.
Abstract: Obtaining labeled data in supervised learning is often
difficult and expensive, and thus the trained learning algorithm tends
to be overfitting due to small number of training data. As a result,
some researchers have focused on using unlabeled data which may
not necessary to follow the same generative distribution as the labeled
data to construct a high-level feature for improving performance on
supervised learning tasks. In this paper, we investigate the impact of
the relationship between unlabeled and labeled data for classification
performance. Specifically, we will apply difference unlabeled data
which have different degrees of relation to the labeled data for
handwritten digit classification task based on MNIST dataset. Our
experimental results show that the higher the degree of relation
between unlabeled and labeled data, the better the classification
performance. Although the unlabeled data that is completely from
different generative distribution to the labeled data provides the lowest
classification performance, we still achieve high classification performance.
This leads to expanding the applicability of the supervised
learning algorithms using unsupervised learning.
Abstract: An additive fuzzy system comprising m rules with
n inputs and p outputs in each rule has at least t m(2n + 2 p + 1)
parameters needing to be tuned. The system consists of a large
number of if-then fuzzy rules and takes a long time to tune its
parameters especially in the case of a large amount of training data
samples. In this paper, a new learning strategy is investigated to cope
with this obstacle. Parameters that tend toward constant values at the
learning process are initially fixed and they are not tuned till the end
of the learning time. Experiments based on applications of the
additive fuzzy system in function approximation demonstrate that the
proposed approach reduces the learning time and hence improves
convergence speed considerably.
Abstract: This paper introduces new algorithms (Fuzzy relative
of the CLARANS algorithm FCLARANS and Fuzzy c Medoids
based on randomized search FCMRANS) for fuzzy clustering of
relational data. Unlike existing fuzzy c-medoids algorithm (FCMdd)
in which the within cluster dissimilarity of each cluster is minimized
in each iteration by recomputing new medoids given current
memberships, FCLARANS minimizes the same objective function
minimized by FCMdd by changing current medoids in such away
that that the sum of the within cluster dissimilarities is minimized.
Computing new medoids may be effected by noise because outliers
may join the computation of medoids while the choice of medoids in
FCLARANS is dictated by the location of a predominant fraction of
points inside a cluster and, therefore, it is less sensitive to the
presence of outliers. In FCMRANS the step of computing new
medoids in FCMdd is modified to be based on randomized search.
Furthermore, a new initialization procedure is developed that add
randomness to the initialization procedure used with FCMdd. Both
FCLARANS and FCMRANS are compared with the robust and
linearized version of fuzzy c-medoids (RFCMdd). Experimental
results with different samples of the Reuter-21578, Newsgroups
(20NG) and generated datasets with noise show that FCLARANS is
more robust than both RFCMdd and FCMRANS. Finally, both
FCMRANS and FCLARANS are more efficient and their outputs
are almost the same as that of RFCMdd in terms of classification
rate.