Alexios Kotsifakos

A survey of query-by-humming similarity methods

Performing similarity search in large databases is a problem of particular interest in many communities, such as music, database, and data mining. Although several solutions have been proposed in the literature that perform well in many application domains, there is no best method to solve this kind of problem in a Query-By-Humming (QBH) application. In QBH the goal is to find the song(s) most similar to a hummed query in an efficient manner. In this paper, we focus on providing a brief overview of the representations to encode music pieces, and also on the methods that have been proposed for QBH or other similarly defined problems.

Query-sensitive distance measure selection for time series nearest neighbor classification

Many distance or similarity measures have been proposed for time series similarity search. However, none of these measures is guaranteed to be optimal when used for 1-Nearest Neighbor (NN) classification. In this paper we study the problem of selecting the most appropriate distance measure, given a pool of time series distance measures and a query, so as to perform NN classification of the query. We propose a framework for solving this problem, by identifying, given the query, the distance measure most likely to produce the correct classification result for that query. From this proposed framework, we derive three specific methods, that differ from each other inthe way they estimatethe probability that a distance measure correctly classifiesa query object. In our experiments, our pool of measures consists of Dynamic Time Warping (DTW), Move-Split-Merge (MSM), and Edit distance with Real Penalty (ERP). Based on experimental evaluation with 45 datasets, the best-performing of the three proposed methods provides the best results in terms of classification error rate, compared to the competitors, which include using the Cross Validation method for selecting the distance measure in each dataset, as well as using a single specific distance measure (DTW, MSM, or ERP) across all datasets.

Model-based search in large time series databases

An important theoretical topic in assistive environments is reasoning about temporal patterns, that represent the sequential output of various sensors, and that can give us information about the health and activities of humans and the state of the environment. The recent growth in the quantity and quality of sensors for assistive environments has made it possible to create large databases of temporal patterns, that store sequences of observations obtained from such sensors over large time intervals. A topic of significant interest is being able to search such large databases so as to identify content of interest, for example activities of a certain type, or information about a patients well-being. In this paper, we study two different approaches for conducting such searches: an exemplar-based approach, where we describe what we are looking for by giving an example, and a model-based approach, where we describe what we are looking for via a generative model. In particular, we describe the two different approaches, and we identify some important pros and cons for each approach. We also perform a comparative evaluation of exemplar-based search using dynamic time warping (DTW), and model-based search using Hidden Markov Models (HMMs), on large real datasets. In our experiments, when the number of training objects per model is sufficiently high, model-based search using HMMs produces more accurate search results than exemplar-based search using DTW.

Deploying In-Network Data Analysis Techniques in Sensor Networks

Sensor Networks have received considerable attention recently, as they provide manifold benefits. Not only are they a means for data acquisition and monitoring of unexplored or inaccessible areas, they are also a low-cost alternative for sensing the environment, which greatly aids to better understand our surroundings. A major motivation in either occasion is to acknowledge endangering situations and take action(s) accordingly. To this end, we would like to enable data mining or analysis techniques on top or, even better, within such networks, due to the prohibitive cost of communication in this setting. In this work, we demonstrate running data mining algorithms on a set of sensors, which are of low-processing power. In addition to showcasing the execution of data analysis algorithms on resource-constrained hardware, our demo is intended to show how to take advantage of the properties of each algorithm to make better use of the sensors and their capabilities. We support the execution and monitoring of these algorithms with a graphical user interface (GUI).

Embedding-based subsequence matching with gaps---range---tolerances: a Query-By-Humming application

We present a subsequence matching framework that allows for gaps in both query and target sequences, employs variable matching tolerance efficiently tuned for each query and target sequence, and constrains the maximum matching range. Using this framework, a dynamic programming method is proposed, called SMBGT, that, given a short query sequence Q and a large database, identifies in quadratic time the subsequence of the database that best matches Q. SMBGT is highly applicable to music retrieval. However, in Query-By-Humming applications, runtime is critical. Hence, we propose a novel embedding-based approach, called ISMBGT, for speeding up search under SMBGT. Using a set of reference sequences, ISMBGT maps both Q and each position of each database sequence into vectors. The database vectors closest to the query vector are identified, and SMBGT is then applied between Q and the subsequences that correspond to those database vectors. The key novelties of ISMBGT are that it does not require training, it is query sensitive, and it exploits the flexibility of SMBGT. We present an extensive experimental evaluation using synthetic and hummed queries on a large music database. Our findings show that ISMBGT can achieve speedups of up to an order of magnitude against brute-force search and over an order of magnitude against cDTW, while maintaining a retrieval accuracy very close to that of brute-force search.

DRESS: dimensionality reduction for efficient sequence search

Similarity search in large sequence databases is a problem ubiquitous in a wide range of application domains, including searching biological sequences. In this paper we focus on protein and DNA data, and we propose a novel approximate method method for speeding up range queries under the edit distance. Our method works in a filter-and-refine manner, and its key novelty is a query-sensitive mapping that transforms the original string space to a new string space of reduced dimensionality. Specifically, it first identifies the

A subsequence matching with gaps-range-tolerances framework: a query-by-humming application

t

Genre classification of symbolic music with SMBGT

t most frequent codewords in the query, and then uses these codewords to convert both the query and the database to a more compact representation. This is achieved by replacing every occurrence of each codeword with a new letter and by removing the remaining parts of the strings. Using this new representation, our method identifies a set of candidate matches that are likely to satisfy the range query, and finally refines these candidates in the original space. The main advantage of our method, compared to alternative methods for whole sequence matching under the edit distance, is that it does not require any training to create the mapping, and it can handle large query lengths with negligible losses in accuracy. Our experimental evaluation demonstrates that, for higher range values and large query sizes, our method produces significantly lower costs and runtimes compared to two state-of-the-art competitor methods.