A preliminary school schedule:
16.09.2013, Mon | 17.09.2013, Tue | 18.09.2013, Wed | 19.09.2013, Thu | 20.09.2013, Fri | 21.09.2013, Sat | ||||||
9:30-11:00 | SCR Small concert hall (UNICS) | SCR Small concert hall (UNICS) | SCR Small concert hall (UNICS) | SCR Small concert hall (UNICS) | Hackathon | ||||||
11:30-13:00 | IIRM ¹104 (Lib) | CCMSR ¹218 (Lib) | IIRM ¹104 (Lib) | CCMSR ¹218 (Lib) | IIRM ¹104 (Lib) | CCMSR ¹218 (Lib) | IIRM ¹104 (Lib) | CCMSR ¹101 (Lib) | IIRM ¹104 (Lib) | CCMSR ¹101 (Lib) | |
14:00-15:30 | LSIR ¹104 (Lib) | QBSH&AFP ¹218 (Lib) | LSIR ¹104 (Lib) | QBSH&AFP ¹218 (Lib) | LSIR ¹104 (Lib) | QBSH&AFP ¹218 (Lib) | LSIR ¹104 (Lib) | QBSH&AFP ¹101 (Lib) | LSIR ¹104 (Lib) | QBSH&AFP ¹101 (Lib) | |
16:00-17:30 | NRMTC ¹104 (Lib) | AAMR ¹218 (Lib) | NRMTC ¹104 (Lib) | AAMR ¹218 (Lib) | NRMTC ¹104 (Lib) | AAMR ¹218 (Lib) | NRMTC ¹104 (Lib) | AAMR ¹101 (Lib) | NRMTC ¹104 (Lib) | AAMR ¹101 (Lib) | |
18:00-19:30 | Yandex ¹109 (Building II) | YSC ¹109 (Building II) | YSC ¹109 (Building II) | Mail.ru ¹109 (Building II) | SCR ¹109 (Building II) | ||||||
welcome party 20:00 | sport evening 20:00 | RuSSIR party 20:30 |
Voice and Music Information Retrieval:
Spoken Content Retrieval: Challenges, Techniques and Applications
Gareth Jones
Spoken Content Retrieval is concerned with searching for relevant spoken content from within archives of speech or multimedia data. The key challenges raised by speech search are the indexing of the spoken content through an appropriate process of speech recognition and efficiently accessing specific content elements from within spoken data. The specific characteristics of speech search in terms of vocabulary and recognition accuracy mean that effective speech search often does not reduce to application of text information retrieval techniques to speech recognition transcripts generated by commercial speech recognition engines. Although text information retrieval techniques are clearly helpful, spoken content retrieval often involves confronting issues such as high levels of noise in the indexed data, a lack of a clearly defined unit of retrieval, and the need to design suitable interfaces for user interaction with temporal spoken content.
This course will introduce the challenges and technologies of spoken content retrieval. It will review the history of spoken data search to date, its component technologies (including an introduction to speech recognition), its relationship to text information retrieval, critical system design issues, domains of application, and issues of interaction with spoken content to support efficient access to relevant material. It will also overview initiatives for evaluation of speech retrieval including: TREC SDR, CLEF CL-SR and MediaEval Search and Hyperlinking, available resources to support research and development in the area of spoken content retrieval, and open challenges going forward.
Slides: S&H part 1 part 2 part 3 part 4 part 5
Content- and Context-based Music Similarity and Retrieval
Markus Schedl & Peter Knees
The course will give participants a solid understanding of basic methodology used in and typical applications of music information retrieval (MIR). In this vein, the lectures will provide a sound and comprehensive, nevertheless easy-to-understand, introduction to selected topics in MIR with a focus on similarity computation and retrieval.
In particular, methods that extract features (i) from the music content by employing audio analysis techniques and (ii) from music context data via web and social media mining techniques will be presented. Based on these features, participants will learn methods to compute similarities between songs and between music artists, a key ingredient of music retrieval systems. The presented approaches are highly valuable for MIR tasks, such as automated music playlist generation, personalized web radio, music recommendation systems, and intelligent user interfaces to music. Another aspect of the class will be on evaluating MIR systems beyond the traditional IR-related measures and the difficulties entailed by the need for objective quantification.
The lectures will be accompanied by a practical exercise, in which participants will learn how to build a music retrieval system that combines music content (audio) and music context (web) information.
Slides: part 1 part 2 part 3 part 4 part 5 part 6
Query by Singing/Humming and Audio Fingerprinting as Two Successful Paradigms of Music Information Retrieval
Jyh-Shing Roger Jang
The course will introduce two of the most successful paradigms of content-based music information retrieval: QBSH (Query by Singing/Humming) and AFP (Audio Fingerprinting). These two paradigms have been successfully used in various companies (Shazam, Soundhound, Intonow, to name a few) to create innovation and fun applications over mobile devices. Our goal is to cover each single step of the retrieval process, such that the audience will gain a fully understand of the whole procedure. These steps include feature extraction, comparison methods, speedup techniques, optimization strategies, etc.
Moreover, we shall discuss the tradeoffs between efficiency and effectiveness of the retrieval, considering the large-scale databases and limited computing power. The issue of implementation over high-computing GPU will also be address. After the course, the audience should have a comprehensive notation of how to create QBSH and AFP prototypes for their own applications.
Slides
Adaptivity in Audio and Music Retrieval
Andreas Nürnberger & Sebastian Stober
This course provides an overview and in-depth discussion of selected ideas and concepts for the adaptation of systems to search and explore audio collections. We first provide, based on typical retrieval scenarios, a clear motivation and a better understanding of the benefits of a user-centered design and personalization. Then we discuss selected approaches that allow to structure and visualize sets of audio objects using user specific interests in more detail. We focus especially on approaches that adapt the underlying similarity measures, which are a central issue in Music Information Retrieval (MIR): In the classical retrieval scenario, similarity is used as an estimate for relevance to rank. Further applications comprise the sorting and organization of music collections by grouping similar objects or generating maps for collection overview. Finally, recommender systems that follow the popular “find more like..."-idea often employ a similarity-based strategy as well. However, music similarity is not a simple concept. For a comparison of music pieces many interrelated features and facets have to be considered. Their individual importance and how they should be combined depends very much on the user and his specific retrieval task. Users of MIR systems may have a varying background and experience music objects like images or music pieces in different ways. Consequently, when comparing, e.g. musical pieces with each other, opinions may diverge. A musician, for instance, might especially look after structures, harmonics or instrumentation (possibly paying conscious- or unconsciously special attention to his own instrument), while non-musicians will perhaps focus more on overall timbre or general mood. Apart from considering individual users, similarity measures also should be tailored for their specific retrieval task to improve the performance of the retrieval system. Throughout the course, prototypical implementations of systems for personalized access to audio and music collections are presented.
Slides: part 1 part 2 part 3 part 4 part 5
General Information retrieval:
Introduction to Information Retrieval Models
Massimo Melucci
This course is an introduction to the IR models. The main thrust of the course is to show the common theoretical framework of the main vectorial and probabilistic models.
To this aim, we will introduce the notions underlying the theory of vector spaces. In particular, the relationship between the notions of vector basis and inner product and, respectively, the notions of random variable and probability will be illustrated. Probability space and random variable will be explained in a way to make the connection to the respective notions of the abstract vector spaces explicit. We will also introduce the notion of statistical correlation in the way it is applied in IR for measuring relationships between information objects. Within this framework, the main retrieval models and methods of latent semantic analysis will also be illustrated; in particular, the vector-space model, the classical probabilistic model, and the language models will be addressed within the theoretical framework.
Techniques for Large Scale Information Retrieval
Paolo Boldi
The finality of this course is to show the students the impact of data size on the design of algorithms and data structures. In fact, the amount of data to be processed makes it impossible to apply standard approaches for their handling, and calls for modern techniques that are scalable, have a low memory footprint and are apt at being easily parallelized and distributed. More often than not, such techniques are approximated and require sophisticated probabilistic tricks for their design and analysis.
The course aims at collecting some important topics related to the analysis of big data, with a twofold aim: on one hand, each such topic is important per se, and is of uttermost importance in building a modern IR search engine; on the other hand, they can be seen as examples of how algorithmic techniques must be adapted to fit the new needs dictated by the presence of large datasets.
Slides: part 1 part 2 part 3 part 4 part 5 part 6
Novel representations and methods in text classification
Manuel Montes-y-Gómez & Hugo Jair Escalante
Two core components of any classification system are the adopted representation for documents and the classification model itself. This tutorial deals with recent advances and developments on both components. The default representation for documents in text classification is the bag-of-words(BOW), where weighting schemes similar to those used in information retrieval are adopted.Whereas this representation has proven to be very helpful for thematic text classification, in novel, non-thematic text classification problems (e.g., authorship attribution, sentiment analysis and opinion mining, etc.), the standard BOW can be outperformed by other advanced representations.
This course is focused on three document representations that have proved to be useful for capturing more information than the raw occurrence of terms in documents as in BOW. The considered representations are: locally weighted BOW, distributional term representations,concise representations and graph-based representations. Likewise, the tutorial covers recent developments in the task of building classification models. Specifically, we consider contextual classification techniques and full model selection methods. The former approach is focused in the design of classifiers that consider the neighborhood of a document for making better predictions.
The latter formulation focuses in the development of automatic methods for building classification systems, that is, black box tools that receive as input a data set and return a very effective classification model.
Slides: part 1 part 2 part 3 part 4 part 5
Workshop on the Tomita Parser: An Instrument for Extracting Facts from Text
Natalia Ostapuk, Tatiana Lando, Sergey Zubkov
At the workshop, we will talk about the Tomita Parser, a text mining tool based on a GLR algorithm created by Masaru Tomita, and demonstrate how to use it using simple examples.
The Tomita Parser is a natural language processing tool created at Yandex. It allows structured data to be extracted from text written in natural language by using context-free grammars and thesauruses. It is used in preparing data for various Yandex services, like Yandex.News and Yandex.Jobs.
The tool is available to all those who wish to use it. You can download it, write your own grammars, and apply it in your projects.
Please install the program ahead of time and make sure that it launches and works. To do so, download the archive file with the program at this link, read the readme.txt file, and perform the actions described there.
Should any problems arise, write to us using the feedback form and we will help you.
Important: don’t forget to bring your laptops with you.
Slides
The crawler
Alexey Voropaev
"Running a web crawler is a challenging task", Brin and Page said in 1998. And this statement is still true. And moreover this topic is still not well covered by literature in contrast to other IR topics. I will show construction of the crawler in our search engine and will explain related problems.
Slides