Browsing by Subject "Language model"

Now showing 1 - 3 of 3

Open Access
The effect of gender bias on hate speech detection
(Springer, 2022-10-08) Şahinuç, F.; Yılmaz, E. H.; Toraman, Ç.; Koç, Aykut
Hate speech against individuals or communities with different backgrounds is a major problem in online social networks. The domain of hate speech has spread to various topics, including race, religion, and gender. Although there are many efforts for hate speech detection in different domains and languages, the effects of gender identity are not solely examined in hate speech detection. Moreover, hate speech detection is mostly studied for particular languages, specifically English, but not low-resource languages, such as Turkish. We examine gender identity-based hate speech detection for both English and Turkish tweets. We compare the performances of state-of-the-art models using 20 k tweets per language. We observe that transformer-based language models outperform bag-of-words and deep learning models, while the conventional bag-of-words model has surprising performances, possibly due to offensive or hate-related keywords. Furthermore, we analyze the effect of debiased embeddings for hate speech detection. We find that the performance can be improved by removing the gender-related bias in neural embeddings since gender-biased words can have offensive or hateful implications.
Open Access
Online nonlinear modeling for big data applications
(2017-12) Khan, Farhan
We investigate online nonlinear learning for several real life, adaptive signal processing and machine learning applications involving big data, and introduce algorithms that are both e cient and e ective. We present novel solutions for learning from the data that is generated at high speed and/or have big dimensions in a non-stationary environment, and needs to be processed on the y. We speci cally focus on investigating the problems arising from adverse real life conditions in a big data perspective. We propose online algorithms that are robust against the non-stationarities and corruptions in the data. We emphasize that our proposed algorithms are universally applicable to several real life applications regardless of the complexities involving high dimensionality, time varying statistics, data structures and abrupt changes. To this end, we introduce a highly robust hierarchical trees algorithm for online nonlinear learning in a high dimensional setting where the data lies on a time varying manifold. We escape the curse of dimensionality by tracking the subspace of the underlying manifold and use the projections of the original high dimensional regressor space onto the underlying manifold as the modi ed regressor vectors for modeling of the nonlinear system. By using the proposed algorithm, we reduce the computational complexity to the order of the depth of the tree and the memory requirement to only linear in the intrinsic dimension of the manifold. We demonstrate the signi cant performance gains in terms of mean square error over the other state of the art techniques through simulated as well as real data. We then consider real life applications of online nonlinear learning modeling, such as network intrusions detection, customers' churn analysis and channel estimation for underwater acoustic communication. We propose sequential and online learning methods that achieve signi cant performance in terms of detection accuracy, compared to the state-of-the-art techniques. We speci cally introduce structured and deep learning methods to develop robust learning algorithms. Furthermore, we improve the performance of our proposed online nonlinear learning models by introducing mixture-of-experts methods and the concept of boosting. The proposed algorithms achieve signi cant performance gain over the state-ofthe- art methods with signi cantly reduced computational complexity and storage requirements in real life conditions.
Open Access
Optimization and machine learning in MRI: applications in rapid MR image reconstruction and encoding models of cortical representations
(2020-02) Shahdloo, Mohammad
Magnetic Resonance Imaging (MRI) is a non-invasive medical imaging modality that is widely used by clinicians and researchers to picture body anatomy and neuronal function. However, long scan time remains a major problem. Recently, multiple techniques have emerged that reduce the acquired MRI signal samples, hence dramatically accelerating the acquisition. These techniques involve sophisticated signal reconstruction procedures that in essence require solving regularized optimization problems, and clinical adoption of accelerated MRI critically relies on self-tuning solutions for these problems. Further to this, recent experimental approaches in cognitive neuroscience favor employing naturalistic audio-visual stimuli that closely resemble humans’ daily-life experience. Yet, these modern paradigms inevitably lead to huge functional MRI (fMRI) datasets that require advanced statistical and computational techniques to uncover the large amount of embedded information. Here, we propose a novel eﬃcient datadriven self-tuning reconstruction method for accelerated MRI. We demonstrate superior performance of the proposed method across various simulated and in vivo datasets and under various scan conﬁgurations. Furthermore, we develop statistical analysis tools to investigate the neural representation of hundreds of action categories in natural movies in the brain via fMRI, and study their attentional modulations. Finally, we develop a model-based framework to estimate temporal extent of semantic information integration in the brain, and investigate its attentional modulations using fMRI data recorded during natural story listening. In short, the methodological and analytical approaches introduced in this thesis greatly beneﬁt clinical utility of accelerated MRI, and enhance our understanding of brain function in daily life.