previous pauseresume next

NEWS

BU/CmpE team wins the Physical Load Sub-Challenge Award at INTERSPEECH Computational Paralinguistics Challenge

Our students Heysem Kaya and Tuğçe Özkaptan under the supervision of Albert Ali Salah and Fikret Gurgen won the Pysical Load Sub-challenge of the Computational Paralinguistics Read more...

Pages

CmpE Events

Today

  1. MS Thesis Defense - Atıf Emre Yüksel - Hate Speech Detection in Turkish News Using a Transformer-based Model Enhanced with Linguistic Features
    • Start time: 10:00am, Wednesday, August 10th
    • End time: 11:00am, Wednesday, August 10th
    • Where: Zoom Meeting
    • Hate speech directed at ethnicities, nationalities, religious identities, and specific groups has increased not only in social media, but also in print media. This creates a need for automated hate speech detection systems that can quickly review and filter print media content before it is provided to readers if it contains hate speech. However, most of the existing automatic hate speech detection models are limited to detecting hate speech without considering the hate speech target group-specific discourse that is often used in news articles. Moreover, there are few datasets that include Turkish print media articles in the hate speech domain. In this study, a new BERT -based model enriched with a set of target-oriented linguistic features for hate speech detection is proposed. The effects of weighting different BERT hidden vectors are also investigated, instead of using only the first hidden vector of the BERT -encoder, which is the classical approach. New BERT -based models that integrate different attention techniques are proposed for combining hidden vectors. A new preprocessed Turkish dataset for hate speech is also published, in which the target group for all hate speech articles is annotated. Experiments on a comprehensive Turkish dataset of news articles labeled for hate speech show that competitive performance in terms of accuracy and F1-score is achieved compared to previous approaches.

    • View this event in Google Calendar
  2. PhD Thesis Defense - Ahmet Alp Kındıroğlu - Transfer Learning for Sign Language Recognition
    • Start time: 10:00am, Wednesday, August 10th
    • End time: 12:00pm, Wednesday, August 10th
    • Where: Zoom Meeting
    • Sign language is a visual language that conveys meaningthrough the arrangement and movement of hands, arms, and facial expressions.Computer vision-based sign language recognition (SLR) can assist in bridgingthe communication gap between hearing and deaf people. SLR is an umbrella termfor a variety of tasks, including isolated sign recognition, sign spotting,continuous sign language recognition, and sign language translation. Researchon sign language recognition (SLR) has made significant progress but relies onvast amounts of data to model and recognize signs. It has not yet generatedviable applications that can do translations for everyday users despite thefact that a substantial amount of effort is being devoted to generating bigannotated sign language datasets for sign languages. In addition, most SLRresearch is focused on a few popular sign languages, leaving the majority ofsign languages, including Turkish Sign Language (TID), as under-resourcedlanguages for developing sign language technologies.In this dissertation, we have highlighted a number of openresearch questions about the development of sign language recognitiontechnologies for TID and have approached the topic from a number of differentangles. We generated BosphorusSign22k, an isolated SLR dataset for TIDcontaining 22k videos from 744 different classes and provided benchmark resultsusing state-of-the-art approaches on this dataset. In order to efficientlymodel signs, we proposed aligned temporal accumulative features (ATAF) as afeature capable of representing isolated sign language gestures as dynamic andstatic subunits. Combined with methods using other modalities, the methodachieves state-of-the-art performance on the BosphorusSign22k dataset. Next, weutilized regularized regression-based multi-task learning methods and proposeda sign language alignment method called task-aware canonical time warping forisolated sign language recognition. The method aimed to align and group signsso as to minimize discrepancies between signs from different sources whileemphasizing differences in signs from different classes. Finally, we utilizedseveral data sources for training isolated sign language recognition models inorder to improve recognition performance for under-resourced languages. Weestablished a benchmark for cross-dataset transfer learning using two existingpublic Turkish SLR datasets and evaluated five supervised transfer learningalgorithms using a temporal graph convolution-based sign language recognitionmethod. Experiments with closed-set and partial-set cross-dataset transferlearning reveal a substantial improvement over combined training andfine-tuning-based baseline techniques.

    • View this event in Google Calendar
  3. PhD Thesis Defense - Serdar Metin - Design, Implementation, and Analysis of Fair Faucets for Blockchain Ecosystems
    • Start time: 01:00pm, Wednesday, August 10th
    • End time: 03:00pm, Wednesday, August 10th
    • Where: AVS Conference Room@BM & Zoom
    • The present dissertation addresses the problem of fairly distributing shared resources in non-commercial blockchain networks. Blockchains are distributed systems that order and timestamp records of a given network of users, in a public, cryptographically secure, and consensual way. The records, which may in kind be events, transaction orders, sets of rules for structured transactions etc. are placed within well-defined data structures called blocks, and they are linked to each other by the virtue of cryptographic pointers, in a total ordering which represents their temporal relations of succession. The ability to operate on the blockchain, and/or to contribute a record to the content of a block are shared resources of the blockchain systems. In commercial networks, these resources are exchanged in return for fiat money, and consequently, fairness is not a relevant problem in terms of computer engineering. In non-commercial networks, however, monetary solutions are not available, by definition. The present non-commercial blockchain networks (e.g. test networks such as Ropsten or Rinkeby, or academic networks such as Bloxberg) employ trivial distribution mechanisms called faucets, which offer fixed amounts of free tokens (called cryptocurrencies) specific to the given network. This mechanism, although simple and efficient, is prone to denial of service (DoS) attacks and cannot address the fairness problem. In the present dissertation, the faucet mechanism is adapted for fair distribution, in line with Max-min Fairness scheme. In total, we contributed six distinct Max-min Fair algorithms as efficient blockchain faucets. The algorithms we contribute are resistant to DoS attacks, low-cost in terms of blockchain computation economics, and they also allow for different user weighting policies. While four of the contributed algorithms provide scalability to unlimited number of users, two of them account for both short term and long term fairness.

    • View this event in Google Calendar
  4. PhD Thesis Defense - Gökhan Çapan - Algorithms for learning from online human behavior and human interaction with learning algorithms
    • Start time: 04:30pm, Wednesday, August 10th
    • End time: 06:30pm, Wednesday, August 10th
    • Where: Zoom Meeting
    • In modern digital systems, algorithms that deliver personalized content shape the user experience and affect user satisfaction, hence long-term engagement with the system. What the system presents also influences the parties providing content to the system since visibility to the user is vital for reachability. Such algorithms learn to deliver personalized content using data on previous user behavior, e.g., their choices, clicks, ratings, etc., interpreted as a proxy for user preferences. In the first part of this work, we review prevalent models for learning from user feedback on items, including our contributions to the literature. As such data is ever-growing, we discuss computational aspects of learning algorithms and focus on software libraries for scalable implementations, including our contributions. The second part is on learning from user interactions with algorithmic personalization systems. Albeit helpful, human behavior is subject to cognitive biases, and data sets comprising their item choices are subject to sampling biases, posing problems to learning algorithms that rely on such data. As users interact with the system, the problem worsens—the algorithms use biased data to compose future content. Further, the algorithms self-reinforce their inaccurate beliefs on user preferences. We review some of the biases and investigate a particular one: the user’s tendency to choose from the alternatives presented by the system, showing the least effort to explore further. To account for it, we develop a Bayesian choice model that explicitly incorporates in the inference of user preferences their limited exposure to a systematically selected subset of items by an algorithm. The model leads to an online learning algorithm of user preferences through interactions.

    • View this event in Google Calendar

Tomorrow

  1. PhD Thesis Defense - Niaz Chalabianloo - Stress Measurement and Regulation in Real-Life Using Affective Technologies
    • Start time: 10:00am, Thursday, August 11th
    • End time: 12:00pm, Thursday, August 11th
    • Where: AVS Conference Room, BM
    • Stress has become one of the main contributors to serious mental and physical health issues in today's world. Existing works in the literature have used Psychophysiological measures and proposed numerous mechanisms to detect stress and administer feedback to help users regulate it. Unobtrusive wearables' popularity is increasingly growing, intertwined with digital health notions, making them efficient, inexpensive, and easily accessible affective self-help technologies. This thesis first aims to investigate and implement stress detection mechanisms in the laboratory and everyday environments using unobtrusive wearable devices. In this regard, we investigate various scenarios, such as how to use laboratory data to improve the results of a daily life scenario. We also explore how adding contextual information such as physical activity and weather information can improve the results. Moreover, we study low-cost and practical methods for emotional regulation in stressful conditions of everyday life. In the next step, a mixed-methods study is conducted. For this, signals from multiple wearables and users' subjective opinions regarding different aspects of wearability were analyzed quantitatively and qualitatively. The next step is an in-depth study in cooperation with HCI researchers, in which we demonstrate the effects of haptic feedback on emotion regulation. As a next step for helping users choose the right device, we evaluate several wearables under completely identical conditions to compare the stress detection quality in wearables with different technologies. Finally, we utilize Explainable AI to make our models more understandable for the end users, and in particular for the psychology and clinical experts. The results of our studies indicate that an integrated detection, notification, and intervention cycle is required to ensure a reliable system for regulating stress in daily life.

    • View this event in Google Calendar

CmpE Announcements

.

Contact us

Department of Computer Engineering, Boğaziçi University,
34342 Bebek, Istanbul, Turkey

  • Phone: +90 212 359 45 23/24
  • Fax: +90 212 2872461
 

Connect with us

We're on Social Networks. Follow us & get in touch.