-
On the Encoding of Gender in Transformer-based ASR Representations
Authors:
Aravind Krishnan,
Badr M. Abdullah,
Dietrich Klakow
Abstract:
While existing literature relies on performance differences to uncover gender biases in ASR models, a deeper analysis is essential to understand how gender is encoded and utilized during transcript generation. This work investigates the encoding and utilization of gender in the latent representations of two transformer-based ASR models, Wav2Vec2 and HuBERT. Using linear erasure, we demonstrate the…
▽ More
While existing literature relies on performance differences to uncover gender biases in ASR models, a deeper analysis is essential to understand how gender is encoded and utilized during transcript generation. This work investigates the encoding and utilization of gender in the latent representations of two transformer-based ASR models, Wav2Vec2 and HuBERT. Using linear erasure, we demonstrate the feasibility of removing gender information from each layer of an ASR model and show that such an intervention has minimal impacts on the ASR performance. Additionally, our analysis reveals a concentration of gender information within the first and last frames in the final layers, explaining the ease of erasing gender in these layers. Our findings suggest the prospect of creating gender-neutral embeddings that can be integrated into ASR frameworks without compromising their efficacy.
△ Less
Submitted 14 June, 2024;
originally announced June 2024.
-
Self-supervised Adaptive Pre-training of Multilingual Speech Models for Language and Dialect Identification
Authors:
Mohammed Maqsood Shaik,
Dietrich Klakow,
Badr M. Abdullah
Abstract:
Pre-trained Transformer-based speech models have shown striking performance when fine-tuned on various downstream tasks such as automatic speech recognition and spoken language identification (SLID). However, the problem of domain mismatch remains a challenge in this area, where the domain of the pre-training data might differ from that of the downstream labeled data used for fine-tuning. In multi…
▽ More
Pre-trained Transformer-based speech models have shown striking performance when fine-tuned on various downstream tasks such as automatic speech recognition and spoken language identification (SLID). However, the problem of domain mismatch remains a challenge in this area, where the domain of the pre-training data might differ from that of the downstream labeled data used for fine-tuning. In multilingual tasks such as SLID, the pre-trained speech model may not support all the languages in the downstream task. To address this challenge, we propose self-supervised adaptive pre-training (SAPT) to adapt the pre-trained model to the target domain and languages of the downstream task. We apply SAPT to the XLSR-128 model and investigate the effectiveness of this approach for the SLID task. First, we demonstrate that SAPT improves XLSR performance on the FLEURS benchmark with substantial gains up to 40.1% for under-represented languages. Second, we apply SAPT on four different datasets in a few-shot learning setting, showing that our approach improves the sample efficiency of XLSR during fine-tuning. Our experiments provide strong empirical evidence that continual adaptation via self-supervision improves downstream performance for multilingual speech models.
△ Less
Submitted 12 December, 2023;
originally announced December 2023.
-
Information-Theoretic Characterization of Vowel Harmony: A Cross-Linguistic Study on Word Lists
Authors:
Julius Steuer,
Badr Abdullah,
Johann-Mattis List,
Dietrich Klakow
Abstract:
We present a cross-linguistic study that aims to quantify vowel harmony using data-driven computational modeling. Concretely, we define an information-theoretic measure of harmonicity based on the predictability of vowels in a natural language lexicon, which we estimate using phoneme-level language models (PLMs). Prior quantitative studies have relied heavily on inflected word-forms in the analysi…
▽ More
We present a cross-linguistic study that aims to quantify vowel harmony using data-driven computational modeling. Concretely, we define an information-theoretic measure of harmonicity based on the predictability of vowels in a natural language lexicon, which we estimate using phoneme-level language models (PLMs). Prior quantitative studies have relied heavily on inflected word-forms in the analysis of vowel harmony. We instead train our models using cross-linguistically comparable lemma forms with little or no inflection, which enables us to cover more under-studied languages. Training data for our PLMs consists of word lists with a maximum of 1000 entries per language. Despite the fact that the data we employ are substantially smaller than previously used corpora, our experiments demonstrate the neural PLMs capture vowel harmony patterns in a set of languages that exhibit this phenomenon. Our work also demonstrates that word lists are a valuable resource for typological research, and offers new possibilities for future studies on low-resource, under-studied languages.
△ Less
Submitted 9 August, 2023;
originally announced August 2023.
-
Exploring electronic, optical, and phononic properties of MgX (X=C, N, and O) monolayers using first principle calculations
Authors:
Nzar Rauf Abdullah,
Botan Jawdat Abdullah,
Yousif Hussein Azeez,
Vidar Gudmundsson
Abstract:
The electronic, the thermal, and the optical properties of hexagonal MgX monolayers (where X=C, N, and O) are investigated via first principles studies. Ab-initio molecular dynamic, AIMD, simulations using NVT ensembles are performed to check the thermodynamic stability of the monolayers. We find that an MgO monolayer has semiconductor properties with a good thermodynamic stability, while the MgC…
▽ More
The electronic, the thermal, and the optical properties of hexagonal MgX monolayers (where X=C, N, and O) are investigated via first principles studies. Ab-initio molecular dynamic, AIMD, simulations using NVT ensembles are performed to check the thermodynamic stability of the monolayers. We find that an MgO monolayer has semiconductor properties with a good thermodynamic stability, while the MgC and the MgN monolayers have metallic characters. The calculated phonon band structures of all the three considered monolayers shows no imaginary nonphysical frequencies, thus indicating that they all have excellent dynamic stability. The MgO monolayer has a larger heat capacity then the MgC and the MgN monolayers. The metallic monolayers demonstrate optical response in the IR as a consequence of the metal properties, whereas the semiconducting MgO monolayer demonstrates an active optical response in the near-UV region. The optical response in the near-UV is beneficial for nanoelectronics and photoelectric applications. A semiconducting monolayer is a great choice for thermal management applications since its thermal properties are more attractive than those of the metallic monolayer in terms of heat capacity, which is related to the change in the internal energy of the system.
△ Less
Submitted 15 July, 2023;
originally announced July 2023.
-
Optical conductivity enhancement and thermal reduction of BN-codoped MgO nanosheet: Significant effects of B-N atomic interaction
Authors:
Nzar Rauf Abdullah,
Botan Jawdat Abdullah,
Yousif Hussein Azeez,
Chi-Shung Tang,
Vidar Gudmundsson
Abstract:
We investigate the electronic, the thermal, and the optical properties of BN-codoped MgO monolayers taking into account the interaction effects between the B and the N dopant atoms. The relatively wide indirect band gap of a pure MgO nanosheet can be changed to a narrow direct band gap by tuning the B-N attractive interaction. The band gap reduction does not only enhance the optical properties, in…
▽ More
We investigate the electronic, the thermal, and the optical properties of BN-codoped MgO monolayers taking into account the interaction effects between the B and the N dopant atoms. The relatively wide indirect band gap of a pure MgO nanosheet can be changed to a narrow direct band gap by tuning the B-N attractive interaction. The band gap reduction does not only enhance the optical properties, including the absorption spectra and the optical conductivity, but also the most intense peak is shifted from the Deep-UV to the visible light region. The red shifting of the absorption spectra and the optical conductivity are caused by the attractive interaction. In addition, both isotropic and anisotropic characteristics are seen in the optical properties depending on the strength of the B-N attractive interaction. The heat capacity is reduced for the BN-doped MgO monolayer, which can be referred to changes in the bond dissociation energy. The bond dissociation energy decreases as the difference in the electronegativities of the bonded atoms decreases. The lower difference in the electronegativities leads to a weaker endothermic process resulting in reduction of the heat capacity. An ab initio molecular dynamics, AIMD, calculation is utilized to check the thermodynamic stability of the pure and the BN-codoped MgO monolayers. We thus confirm that the BN-codopant atoms can be used to gain control of the properties of MgO monolayers for thermo- and opto-electronic devices.
△ Less
Submitted 15 July, 2023;
originally announced July 2023.
-
Planar buckling controlled optical conductivity of SiC monolayer from Deep-UV to visible light region: A first-principles study
Authors:
Nzar Rauf Abdullah,
Hunar Omar Rashid,
Botan Jawdat Abdullah,
Chi-Shung Tang,
Vidar Gudmundsson
Abstract:
The electrical and optical properties of flat and planar buckled siligraphene (SiC) monolayer are examined using a first principles approach. Buckling between the Si and the C atoms in SiC structures influences and impacts the properties of the 2D nanomaterial, according to our results. The electron density of a planar SiC monolayer is calculated, as well as the effects of buckling on it. Accordin…
▽ More
The electrical and optical properties of flat and planar buckled siligraphene (SiC) monolayer are examined using a first principles approach. Buckling between the Si and the C atoms in SiC structures influences and impacts the properties of the 2D nanomaterial, according to our results. The electron density of a planar SiC monolayer is calculated, as well as the effects of buckling on it. According to our findings, a siligraphene monolayer is a semiconductor nanomaterial with a direct electronic band gap that decreases as the planar buckling rises. The contributions to the density of states differ owing to changes in the system's structure. Another explanation is that planar buckling reduces the sp$^2$ overlap**, breaking the bond symmetry causing it to become a sp$^3$ bond. We show that increased planar buckling between the Si and the C atoms alters the monolayer's optical, mechanical, and thermal properties. A managed planar buckling increases the optical conductivity with a significant shift in the far visible range, as all optical spectra features are red shifted, still remaining visible. Instead of a $σ\text{-}σ$ covalent bond, the sp$^3$ hybridization produces a stronger $σ\text{-}π$ bond. Optical characteristics such as the dielectric function, the absorbance, and the optical conductivity of a SiC monolayer are investigated for both parallel and perpendicular polarization of the incoming electric field for both flat and planar buckled systems. The findings show that the optical properties are influenced for both of these two polarizations, with a significant change in the optical spectrum from the near visible to the far visible. The ability to manipulate the optical and electrical characteristics of this critical 2D material through planar buckling opens up new technological possibilities, especially for optoelectronic devices.
△ Less
Submitted 15 July, 2023;
originally announced July 2023.
-
Buckling effects in AlN monolayers: Shifting and enhancing optical characteristics from the UV to the near visible light range
Authors:
Nzar Rauf Abdullah,
Botan Jawdat Abdullah,
Hunar Omar Rashid,
Vidar Gudmundsson
Abstract:
The structural, electronic, and optical properties of flat and buckled AlN monolayers are investigated using first-principles approaches. The band gap of a flat AlN monolayer is changed from an indirect one to a direct one, when the planar buckling increases, primarily due to diminishing sp$^2$ overlap** and bond symmetry breaking in the conversion to sp$^3$ bonds. The sp$^3$ hybridization thus…
▽ More
The structural, electronic, and optical properties of flat and buckled AlN monolayers are investigated using first-principles approaches. The band gap of a flat AlN monolayer is changed from an indirect one to a direct one, when the planar buckling increases, primarily due to diminishing sp$^2$ overlap** and bond symmetry breaking in the conversion to sp$^3$ bonds. The sp$^3$ hybridization thus results in a stronger $σ\text{-}π$ bond rather than a $σ\text{-}σ$ covalent bond. The calculations of the phonon band structure indicates that the buckled AlN monolayers are structurally and dynamically stable. The optical properties, such as the dielectric function, the refractive index, and the optical conductivity of an AlN monolayer are evaluated for both flat systems and those impacted with planar buckling. The flat AlN monolayer has outstanding optical characteristics in the Deep-UV and absorbs more effectively in the UV spectrum due to its large band gap. The results reveal that optical aspects are enhanced along different directions of light polarization, with a considerable shift in the optical spectrum from Deep-UV into the visible range. Additionally, depending on the polarization direction of the incoming light, increased planar buckling enhances the optical conductivity in both the visible and the Deep-UV domains. The ability to modify the optical and electronic properties of these essential 2D materials using planar buckling technique opens up new technological possibilities, particularly for optoelectronic devices.
△ Less
Submitted 18 July, 2023; v1 submitted 15 July, 2023;
originally announced July 2023.
-
An Information-Theoretic Analysis of Self-supervised Discrete Representations of Speech
Authors:
Badr M. Abdullah,
Mohammed Maqsood Shaik,
Bernd Möbius,
Dietrich Klakow
Abstract:
Self-supervised representation learning for speech often involves a quantization step that transforms the acoustic input into discrete units. However, it remains unclear how to characterize the relationship between these discrete units and abstract phonetic categories such as phonemes. In this paper, we develop an information-theoretic framework whereby we represent each phonetic category as a dis…
▽ More
Self-supervised representation learning for speech often involves a quantization step that transforms the acoustic input into discrete units. However, it remains unclear how to characterize the relationship between these discrete units and abstract phonetic categories such as phonemes. In this paper, we develop an information-theoretic framework whereby we represent each phonetic category as a distribution over discrete units. We then apply our framework to two different self-supervised models (namely wav2vec 2.0 and XLSR) and use American English speech as a case study. Our study demonstrates that the entropy of phonetic distributions reflects the variability of the underlying speech sounds, with phonetically similar sounds exhibiting similar distributions. While our study confirms the lack of direct, one-to-one correspondence, we find an intriguing, indirect relationship between phonetic categories and discrete units.
△ Less
Submitted 4 June, 2023;
originally announced June 2023.
-
Develo** the Reliable Shallow Supervised Learning for Thermal Comfort using ASHRAE RP-884 and ASHRAE Global Thermal Comfort Database II
Authors:
Kanisius Karyono,
Badr M. Abdullah,
Alison J. Cotgrave,
Ana Bras,
Jeff Cullen
Abstract:
The artificial intelligence (AI) system designer for thermal comfort faces insufficient data recorded from the current user or overfitting due to unreliable training data. This work introduces the reliable data set for training the AI subsystem for thermal comfort. This paper presents the control algorithm based on shallow supervised learning, which is simple enough to be implemented in the Intern…
▽ More
The artificial intelligence (AI) system designer for thermal comfort faces insufficient data recorded from the current user or overfitting due to unreliable training data. This work introduces the reliable data set for training the AI subsystem for thermal comfort. This paper presents the control algorithm based on shallow supervised learning, which is simple enough to be implemented in the Internet of Things (IoT) system for residential usage using ASHRAE RP-884 and ASHRAE Global Thermal Comfort Database II. No training data for thermal comfort is available as reliable as this dataset, but the direct use of this data can lead to overfitting. This work offers the algorithm for data filtering and semantic data augmentation for the ASHRAE database for the supervised learning process. Overfitting always becomes a problem due to the psychological aspect involved in the thermal comfort decision. The method to check the AI system based on the psychrometric chart against overfitting is presented. This paper also assesses the most important parameters needed to achieve human thermal comfort. This method can support the development of reinforced learning for thermal comfort.
△ Less
Submitted 3 March, 2023;
originally announced March 2023.
-
Analyzing the Representational Geometry of Acoustic Word Embeddings
Authors:
Badr M. Abdullah,
Dietrich Klakow
Abstract:
Acoustic word embeddings (AWEs) are vector representations such that different acoustic exemplars of the same word are projected nearby in the embedding space. In addition to their use in speech technology applications such as spoken term discovery and keyword spotting, AWE models have been adopted as models of spoken-word processing in several cognitively motivated studies and have been shown to…
▽ More
Acoustic word embeddings (AWEs) are vector representations such that different acoustic exemplars of the same word are projected nearby in the embedding space. In addition to their use in speech technology applications such as spoken term discovery and keyword spotting, AWE models have been adopted as models of spoken-word processing in several cognitively motivated studies and have been shown to exhibit human-like performance in some auditory processing tasks. Nevertheless, the representational geometry of AWEs remains an under-explored topic that has not been studied in the literature. In this paper, we take a closer analytical look at AWEs learned from English speech and study how the choice of the learning objective and the architecture shapes their representational profile. To this end, we employ a set of analytic techniques from machine learning and neuroscience in three different analyses: embedding space uniformity, word discriminability, and representational consistency. Our main findings highlight the prominent role of the learning objective on sha** the representation profile compared to the model architecture.
△ Less
Submitted 8 January, 2023;
originally announced January 2023.
-
Role of planar buckling on the electronic, thermal, and optical properties of Germagraphene nanosheets
Authors:
Nzar Rauf Abdullah,
Yousif Hussein Azeez,
Botan Jawdat Abdullah,
Hunar Omar Rashid,
Andrei Manolescu,
Vidar Gudmundsson
Abstract:
We report the electronic, the thermal, and the optical properties of a Germagraphene (GeC) monolayer taking into account buckling effects. The relatively wide direct band gap of a flat GeC nanosheet can be changed by tuning the planar buckling. A GeC monolayer has an sp$^2$ hybridization in which the contribution of an $s$-orbital is half of the contribution of a $p$-orbital leading to stronger…
▽ More
We report the electronic, the thermal, and the optical properties of a Germagraphene (GeC) monolayer taking into account buckling effects. The relatively wide direct band gap of a flat GeC nanosheet can be changed by tuning the planar buckling. A GeC monolayer has an sp$^2$ hybridization in which the contribution of an $s$-orbital is half of the contribution of a $p$-orbital leading to stronger $σ\text{-}σ$ bonds compared to the $σ\text{-}π$ bonds. Increasing the planar buckling, the contribution of an $s$-orbital is decreased while the contribution of a $p$-orbital is increased resulting in a sp$^3$-hybridization in which the $σ\text{-}π$ bond becomes stronger than the $σ\text{-}σ$ bond. As a result, the band gap of a buckled GeC is reduced and thus the thermal and the optical properties are significantly modified. We find that the heat capacity of the buckled GeC is decreased at low values of planar buckling, which is caused by the anticrossing of the optical and the acoustic phonon modes affecting phonon scattering processes. The resulting optical properties, such as the dielectric function, the refractive index, the electron energy loss spectra, the absorption, and the optical conductivity show that a buckled GeC nanosheet has increased optical activities in the visible light region compared to a flat GeC. The optical conductivity is red shifted from the near ultraviolet to the visible light region, when the planar buckling is increased. We can thus confirm that the buckling can be seen as another parameter to improve GeC monolayers for optoelectronic devices.
△ Less
Submitted 9 October, 2022;
originally announced October 2022.
-
Integrating Form and Meaning: A Multi-Task Learning Model for Acoustic Word Embeddings
Authors:
Badr M. Abdullah,
Bernd Möbius,
Dietrich Klakow
Abstract:
Models of acoustic word embeddings (AWEs) learn to map variable-length spoken word segments onto fixed-dimensionality vector representations such that different acoustic exemplars of the same word are projected nearby in the embedding space. In addition to their speech technology applications, AWE models have been shown to predict human performance on a variety of auditory lexical processing tasks…
▽ More
Models of acoustic word embeddings (AWEs) learn to map variable-length spoken word segments onto fixed-dimensionality vector representations such that different acoustic exemplars of the same word are projected nearby in the embedding space. In addition to their speech technology applications, AWE models have been shown to predict human performance on a variety of auditory lexical processing tasks. Current AWE models are based on neural networks and trained in a bottom-up approach that integrates acoustic cues to build up a word representation given an acoustic or symbolic supervision signal. Therefore, these models do not leverage or capture high-level lexical knowledge during the learning process. In this paper, we propose a multi-task learning model that incorporates top-down lexical knowledge into the training procedure of AWEs. Our model learns a map** between the acoustic input and a lexical representation that encodes high-level information such as word semantics in addition to bottom-up form-based supervision. We experiment with three languages and demonstrate that incorporating lexical knowledge improves the embedding space discriminability and encourages the model to better separate lexical categories.
△ Less
Submitted 18 September, 2022; v1 submitted 14 September, 2022;
originally announced September 2022.
-
Study of the buckling effects on the electrical and optical properties of the group III-Nitride monolayers
Authors:
Nzar Rauf Abdullah,
Botan Jawdat Abdullah,
Hunar Omar Rashid,
Chi-Shung Tang,
Vidar Gudmundsson
Abstract:
We consider electronic and optical properties of group III-Nitride monolayers using first-principle calculations. The group III-Nitride monolayers have flat hexagonal structures with almost zero planar buckling, $Δ$. By tuning the $Δ$, the strong $σ\text{-}σ$ bond through sp$^2$ hybridization of a flat form of these monolayers can be changed to a stronger $σ\text{-}π$ bond through sp$^3$ hybridiza…
▽ More
We consider electronic and optical properties of group III-Nitride monolayers using first-principle calculations. The group III-Nitride monolayers have flat hexagonal structures with almost zero planar buckling, $Δ$. By tuning the $Δ$, the strong $σ\text{-}σ$ bond through sp$^2$ hybridization of a flat form of these monolayers can be changed to a stronger $σ\text{-}π$ bond through sp$^3$ hybridization. Consequently, the band gaps of the monolayers are tuned due to a dislocation of the $s$- and $p$-orbitals towards the Fermi energy. The band gaps decrease with increasing $Δ$ for those flat monolayers, which have a band gap greater than $1.0$ eV, while no noticeable change or a flat dispersion of the band gap is seen for the flat monolayers, that have a band gap less than $1.0$ eV. The decreased band gap causes a decrease in the excitation energy, and thus the static dielectric function, refractive index, and the optical conductivity are increased. In contrast, the flat band gap dispersion of few monolayers in the group III-Nitride induces a reduction in the static dielectric function, the refractive index, and the optical conductivity. We therefore confirm that tuning of the planar buckling can be used to control the physical properties of these monolayers, both for an enhancement and a reduction of the optical properties. These results are of interest for the design of optoelectric devices in nanoscale systems.
△ Less
Submitted 1 July, 2022;
originally announced July 2022.
-
Enhanced ultraviolet absorption in BN monolayers caused by tunable buckling
Authors:
Nzar Rauf Abdullah,
Botan Jawdat Abdullah,
Chi-Shung Tang,
Vidar Gudmundsson
Abstract:
The optical properties of a hexagonal Boron Nitride (BN) monolayer across the UV spectrum are studied by tuning its planar buckling. The strong $σ\text{-}σ$ bond through sp$^2$ hybridization of a flat BN monolayer can be changed to a stronger $σ\text{-}π$ bond through sp$^3$ hybridization by increasing the planar buckling. This gives rise to the $s$- and $p$-orbital contributions to form a density…
▽ More
The optical properties of a hexagonal Boron Nitride (BN) monolayer across the UV spectrum are studied by tuning its planar buckling. The strong $σ\text{-}σ$ bond through sp$^2$ hybridization of a flat BN monolayer can be changed to a stronger $σ\text{-}π$ bond through sp$^3$ hybridization by increasing the planar buckling. This gives rise to the $s$- and $p$-orbital contributions to form a density of states around the Fermi energy, and these states dislocate to a lower energy in the presence of an increased planar buckling. Consequently, the wide band gap of a flat BN monolayer is reduced to a smaller band gap in a buckled BN monolayer enhancing its optical activity in the Deep-UV region. The optical properties such as the dielectric function, the reflectivity, the absorption, and the optical conductivity spectra are investigated. It is shown that the absorption rate can be enhanced by $(12\text{-}15)\%$ for intermediate values of planar buckling in the Deep-UV region, and $(15\text{-}20)\%$ at higher values of planar buckling in the near-UV region. Furthermore, the optical conductivity is enhanced by increased planar buckling in both the visible and the Deep-UV regions depending on the direction of the polarization of the incoming light. Our results may be useful for optoelectronic BN monolayer devices in the UV range including UV spectroscopy, deep-UV communications, and UV photodetectors.
△ Less
Submitted 1 January, 2022;
originally announced January 2022.
-
Electronic and Optical properties of Metallic Nitride: A comparative study between the MN (M=Al, Ga, In, Tl) monolayers
Authors:
Nzar Rauf Abdullah,
Botan Jawdat Abdullah,
Vidar Gudmundsson
Abstract:
The electronic and the optical properties of metallic nitride (MN) monolayers are studied using a DFT formalism. In most of these monolayers, the electron density of the metallic atoms is much higher than that of the nitride atoms, and ionic, covalent, and metallic bonds are found in M-N bonds, resulting in fascinating electronic and optical properties. The optical band gap is varied from almost…
▽ More
The electronic and the optical properties of metallic nitride (MN) monolayers are studied using a DFT formalism. In most of these monolayers, the electron density of the metallic atoms is much higher than that of the nitride atoms, and ionic, covalent, and metallic bonds are found in M-N bonds, resulting in fascinating electronic and optical properties. The optical band gap is varied from almost $0.0$ to $3.0$~eV for the MN monolayers depending on the bond type between the metallic and the nitride atoms, as well as the contribution of the type of orbitals around the Fermi energy. The optical properties such as the dielectric function, the excitation spectra, the refractive index, the reflectivity, and the optical conductivity of MN monolayers are calculated. The excitation energy and static dielectric constant are found to be inversely proportional to the band gap at low photon energy. The MN monolayers with a large band gap have good visible light functionality, while the MN monolayers with a lower band gap are found to be active in the infrared region. Furthermore, it is shown that the optical properties of MN monolayers show a strong anisotropy with respect to the polarization of the incoming light. Consequently, our results for the optical properties of MN monolayers show that they could be beneficial in optoelectronic device applications.
△ Less
Submitted 31 December, 2021;
originally announced January 2022.
-
DFT study of tunable electronic, magnetic, thermal, and optical properties of a Ga$_2$Si$_6$ monolayer
Authors:
Nzar Rauf Abdullah,
Botan Jawdat Abdullah,
Vidar Gudmundsson
Abstract:
The electrical, magnetic, thermal and optical characteristics of Gallium (Ga) doped silicene are investigated using density functional theory (DFT). The effect of do** is studied by tuning dopant concentrations as well as examining varied do** distances, and atomic dopant interactions for the same substitutional do** concentration. The results indicate that the Ga atoms alter the band struct…
▽ More
The electrical, magnetic, thermal and optical characteristics of Gallium (Ga) doped silicene are investigated using density functional theory (DFT). The effect of do** is studied by tuning dopant concentrations as well as examining varied do** distances, and atomic dopant interactions for the same substitutional do** concentration. The results indicate that the Ga atoms alter the band structure and the band gap in the silicene monolayer at various concentrations, which can be referred back to to the repulsive interaction of Ga-Ga atoms. The band gap is determined by the interaction strength of the Ga-Ga atoms, the Coulomb repulsive force, and it does not always widen as do** concentration increases. In addition, our spin-polarized DFT calculations show that these monolayers behave like nonmagnetic semiconductors, exhibiting symmetric spin-up and spin-down channels. The repulsive interaction between the Ga atoms causes a symmetry breaking of the monolayers. As a consequence, a Ga dopant can open the band gap, leading to better thermoelectric properties such as the Seebeck coefficient and the figure of merit, as well as an increase in the optical response. As a result of our estimates, Ga doped silicene monolayers could be advantageous in thermoelectric and optoelectronic devices.
△ Less
Submitted 11 October, 2021;
originally announced October 2021.
-
How Familiar Does That Sound? Cross-Lingual Representational Similarity Analysis of Acoustic Word Embeddings
Authors:
Badr M. Abdullah,
Iuliia Zaitova,
Tania Avgustinova,
Bernd Möbius,
Dietrich Klakow
Abstract:
How do neural networks "perceive" speech sounds from unknown languages? Does the typological similarity between the model's training language (L1) and an unknown language (L2) have an impact on the model representations of L2 speech signals? To answer these questions, we present a novel experimental design based on representational similarity analysis (RSA) to analyze acoustic word embeddings (AWE…
▽ More
How do neural networks "perceive" speech sounds from unknown languages? Does the typological similarity between the model's training language (L1) and an unknown language (L2) have an impact on the model representations of L2 speech signals? To answer these questions, we present a novel experimental design based on representational similarity analysis (RSA) to analyze acoustic word embeddings (AWEs) -- vector representations of variable-duration spoken-word segments. First, we train monolingual AWE models on seven Indo-European languages with various degrees of typological similarity. We then employ RSA to quantify the cross-lingual similarity by simulating native and non-native spoken-word processing using AWEs. Our experiments show that typological similarity indeed affects the representational similarity of the models in our study. We further discuss the implications of our work on modeling speech processing and language similarity with neural networks.
△ Less
Submitted 21 September, 2021;
originally announced September 2021.
-
High thermoelectric and optical conductivity driven by the interaction of Boron and Nitrogen dopant atoms with a 2D monolayer Beryllium Oxide
Authors:
Nzar Rauf Abdullah,
Botan Jawdat Abdullah,
Vidar Gudmundsson
Abstract:
The electronic, thermal and optical properties of a monolayer BeO with Boron (B) and Nitrogen (N) co-dopant atoms are studied by means of a density functional theory computation. Our calculations reveal that BeO with BN-codopant atoms can give rise to more effective and outstanding performance for the thermal and optical responses. More significantly, the monolayer BeO with BN codopant atoms becom…
▽ More
The electronic, thermal and optical properties of a monolayer BeO with Boron (B) and Nitrogen (N) co-dopant atoms are studied by means of a density functional theory computation. Our calculations reveal that BeO with BN-codopant atoms can give rise to more effective and outstanding performance for the thermal and optical responses. More significantly, the monolayer BeO with BN codopant atoms becomes a semiconductor with a direct band gap in comparison with the insulator behavior of pristine BeO. The particular attention of this work is paid to the influence of the atomic configuration and the interaction of the B and N dopant atoms with BeO. The interaction of the B and N atoms with the BeO monolayer diminishes degenerate energy states forming flat bands. It is also found that there is a strong attractive interaction between the O and N atoms forming a strong sigma bond breaking the symmetry of BeO structure. Consequently, the band gap is reduced leading to a semiconductor behavior with improved thermoelectric properties such as the Seebeck coefficient and the figure of merit. The reduced band gap and the flat bands induce a high optical responses such as the refractive index, the reflectivity and the optical conductivity in the visible light region. In addition, the anisotropy of a monolayer BeO with B and N atoms regarding different direction of electromagnetic polarization is presented. We anticipate that our results can be useful for design of both thermoelectric and optoelectronic devices.
△ Less
Submitted 13 September, 2021;
originally announced September 2021.
-
Enhanced electronic and optical responses of Nitrogen- or Boron-doped BeO monolayer: First principle computation
Authors:
Nzar Rauf Abdullah,
Botan Jawdat Abdullah,
Hunar Omar Rashid,
Chi-Shung Tang,
Andrei Manolescu,
Vidar Gudmundsson
Abstract:
In this work, the electronic and optical properties of a Nitrogen (N) or a Boron (B) doped BeO monolayer are investigated in the framework of density functional theory. It is known that the band gap of a BeO monolayer is large leading to poor material for optoelectronic devices in a wide range of energy. Using substitutional N or B dopant atoms, we find that the band gap can be tuned and the optic…
▽ More
In this work, the electronic and optical properties of a Nitrogen (N) or a Boron (B) doped BeO monolayer are investigated in the framework of density functional theory. It is known that the band gap of a BeO monolayer is large leading to poor material for optoelectronic devices in a wide range of energy. Using substitutional N or B dopant atoms, we find that the band gap can be tuned and the optical properties can be improved. In the N(B)-doped BeO monolayer, the Fermi energy slightly crosses the valence(conduction) band forming a degenerate semiconductor structure. The N or B atoms thus generate new states around the Fermi energy increasing the optical conductivity in the visible light region. Furthermore, the influences of dopant atoms on the electronic structure, the stability, the dispersion energy, the density of states, and optical properties such as the plasmon frequency, the excitation spectra, the dielectric functions, the static dielectric constant, and the electron energy loss function are discussed for different directions of polarizations for the incoming electric field.
△ Less
Submitted 29 August, 2021;
originally announced August 2021.
-
Modulation of electronic and thermal proprieties of TaMoS$_2$ by controlling the repulsive interaction between Ta dopant atoms
Authors:
Nzar Rauf Abdullah,
Botan Jawdat Abdullah,
Hunar Omar Rashid,
Chi-Shung Tang,
Vidar Gudmundsson
Abstract:
We theoretically study the electronic and the thermal characteristics of Tantalum, Ta, doped Molybdenum disulfide, MoS$_2$, using density functional theory. It has been shown that the MoS$_2$ monolayer is not a good material for thermoelectric devices due to its relatively large band gap. We find that a Ta doped MoS$_2$ forming a TaMoS$_2$ monolayer can be useful for thermoelectric devices. The pa…
▽ More
We theoretically study the electronic and the thermal characteristics of Tantalum, Ta, doped Molybdenum disulfide, MoS$_2$, using density functional theory. It has been shown that the MoS$_2$ monolayer is not a good material for thermoelectric devices due to its relatively large band gap. We find that a Ta doped MoS$_2$ forming a TaMoS$_2$ monolayer can be useful for thermoelectric devices. The particular attention of this work is paid to the interaction effect between the Ta atoms in the MoS$_2$ structure. We find that the interaction type is repulsive. It introduces an asymmetry in the density of states, DOS, reducing the band gap. In the presence of a strong repulsive interaction of Ta-Ta atoms, new states in the DOS around the Fermi energy are found leading to a reduction of the band gap. Consequently, a high Seebeck coefficient and figure of merit are seen over a wide range of energy around the Fermi energy. In contrast, a small reduction of the band gap and a vanishing degeneracy of the valence and the conduction bands are observed for the case of a weak Ta-Ta repulsive interaction leading to less promising thermoelectric properties.
△ Less
Submitted 8 August, 2021;
originally announced August 2021.
-
Do Acoustic Word Embeddings Capture Phonological Similarity? An Empirical Study
Authors:
Badr M. Abdullah,
Marius Mosbach,
Iuliia Zaitova,
Bernd Möbius,
Dietrich Klakow
Abstract:
Several variants of deep neural networks have been successfully employed for building parametric models that project variable-duration spoken word segments onto fixed-size vector representations, or acoustic word embeddings (AWEs). However, it remains unclear to what degree we can rely on the distance in the emerging AWE space as an estimate of word-form similarity. In this paper, we ask: does the…
▽ More
Several variants of deep neural networks have been successfully employed for building parametric models that project variable-duration spoken word segments onto fixed-size vector representations, or acoustic word embeddings (AWEs). However, it remains unclear to what degree we can rely on the distance in the emerging AWE space as an estimate of word-form similarity. In this paper, we ask: does the distance in the acoustic embedding space correlate with phonological dissimilarity? To answer this question, we empirically investigate the performance of supervised approaches for AWEs with different neural architectures and learning objectives. We train AWE models in controlled settings for two languages (German and Czech) and evaluate the embeddings on two tasks: word discrimination and phonological similarity. Our experiments show that (1) the distance in the embedding space in the best cases only moderately correlates with phonological distance, and (2) improving the performance on the word discrimination task does not necessarily yield models that better reflect word phonological similarity. Our findings highlight the necessity to rethink the current intrinsic evaluations for AWEs.
△ Less
Submitted 16 June, 2021;
originally announced June 2021.
-
SIGTYP 2021 Shared Task: Robust Spoken Language Identification
Authors:
Elizabeth Salesky,
Badr M. Abdullah,
Sabrina J. Mielke,
Elena Klyachko,
Oleg Serikov,
Edoardo Ponti,
Ritesh Kumar,
Ryan Cotterell,
Ekaterina Vylomova
Abstract:
While language identification is a fundamental speech and language processing task, for many languages and language families it remains a challenging task. For many low-resource and endangered languages this is in part due to resource availability: where larger datasets exist, they may be single-speaker or have different domains than desired application scenarios, demanding a need for domain and s…
▽ More
While language identification is a fundamental speech and language processing task, for many languages and language families it remains a challenging task. For many low-resource and endangered languages this is in part due to resource availability: where larger datasets exist, they may be single-speaker or have different domains than desired application scenarios, demanding a need for domain and speaker-invariant language identification systems. This year's shared task on robust spoken language identification sought to investigate just this scenario: systems were to be trained on largely single-speaker speech from one domain, but evaluated on data in other domains recorded from speakers under different recording circumstances, mimicking realistic low-resource scenarios. We see that domain and speaker mismatch proves very challenging for current methods which can perform above 95% accuracy in-domain, which domain adaptation can address to some degree, but that these conditions merit further investigation to make spoken language identification accessible in many scenarios.
△ Less
Submitted 7 June, 2021;
originally announced June 2021.
-
Properties of BC$_6$N monolayer derived by first-principle computation: Influences of interactions between dopant atoms
Authors:
Nzar Rauf Abdullah,
Botan Jawdat Abdullah,
Chi-Shung Tang,
Vidar Gudmundsson
Abstract:
The properties of graphene-like BC$_6$N semiconductor are studied using density functional theory taking into account the attractive interaction between B and N atoms. In the presence of a strong attractive interaction between B and N dopant atoms, the electron charge distribution is highly localized along the B-N bonds, while for a weaker attractive interaction the electrons are delocalized along…
▽ More
The properties of graphene-like BC$_6$N semiconductor are studied using density functional theory taking into account the attractive interaction between B and N atoms. In the presence of a strong attractive interaction between B and N dopant atoms, the electron charge distribution is highly localized along the B-N bonds, while for a weaker attractive interaction the electrons are delocalized along the entire hexagonal ring of BC$_6$N. Furthermore, when both B and N atoms are doped at the same site of the hexagon, the breaking of the sub-lattice symmetry is low producing a small bandgap. In contrast, if the dopant atoms are at different sites, a high sub-lattice symmetry breaking is found leading to a large bandgap. The influences of electron localization/delocalization and the tunable bandgap on thermal behaviors such as the electronic thermal conductivity, the Seebeck coefficient, and the figure of merit, and optical properties such as the dielectric function, the excitation spectra, the refractive index, the electron energy loss spectra, the reflectivity, and the optical conductivity are presented. An enhancement with a red shift of the optical conductivity at low energy range is seen while a reduction at the high energy range is found indicating that the BC$_6$N structure may be useful for optoelectronic devices in the low energy, visible range.
△ Less
Submitted 1 June, 2021;
originally announced June 2021.
-
A Closer Look at Linguistic Knowledge in Masked Language Models: The Case of Relative Clauses in American English
Authors:
Marius Mosbach,
Stefania Degaetano-Ortlieb,
Marie-Pauline Krielke,
Badr M. Abdullah,
Dietrich Klakow
Abstract:
Transformer-based language models achieve high performance on various tasks, but we still lack understanding of the kind of linguistic knowledge they learn and rely on. We evaluate three models (BERT, RoBERTa, and ALBERT), testing their grammatical and semantic knowledge by sentence-level probing, diagnostic cases, and masked prediction tasks. We focus on relative clauses (in American English) as…
▽ More
Transformer-based language models achieve high performance on various tasks, but we still lack understanding of the kind of linguistic knowledge they learn and rely on. We evaluate three models (BERT, RoBERTa, and ALBERT), testing their grammatical and semantic knowledge by sentence-level probing, diagnostic cases, and masked prediction tasks. We focus on relative clauses (in American English) as a complex phenomenon needing contextual information and antecedent identification to be resolved. Based on a naturalistic dataset, probing shows that all three models indeed capture linguistic knowledge about grammaticality, achieving high performance. Evaluation on diagnostic cases and masked prediction tasks considering fine-grained linguistic knowledge, however, shows pronounced model-specific weaknesses especially on semantic knowledge, strongly impacting models' performance. Our results highlight the importance of (a)model comparison in evaluation task and (b) building up claims of model performance and the linguistic knowledge they capture beyond purely probing-based evaluations.
△ Less
Submitted 2 November, 2020;
originally announced November 2020.
-
Rediscovering the Slavic Continuum in Representations Emerging from Neural Models of Spoken Language Identification
Authors:
Badr M. Abdullah,
Jacek Kudera,
Tania Avgustinova,
Bernd Möbius,
Dietrich Klakow
Abstract:
Deep neural networks have been employed for various spoken language recognition tasks, including tasks that are multilingual by definition such as spoken language identification. In this paper, we present a neural model for Slavic language identification in speech signals and analyze its emergent representations to investigate whether they reflect objective measures of language relatedness and/or…
▽ More
Deep neural networks have been employed for various spoken language recognition tasks, including tasks that are multilingual by definition such as spoken language identification. In this paper, we present a neural model for Slavic language identification in speech signals and analyze its emergent representations to investigate whether they reflect objective measures of language relatedness and/or non-linguists' perception of language similarity. While our analysis shows that the language representation space indeed captures language relatedness to a great extent, we find perceptual confusability between languages in our study to be the best predictor of the language representation similarity.
△ Less
Submitted 22 October, 2020;
originally announced October 2020.
-
Cross-Domain Adaptation of Spoken Language Identification for Related Languages: The Curious Case of Slavic Languages
Authors:
Badr M. Abdullah,
Tania Avgustinova,
Bernd Möbius,
Dietrich Klakow
Abstract:
State-of-the-art spoken language identification (LID) systems, which are based on end-to-end deep neural networks, have shown remarkable success not only in discriminating between distant languages but also between closely-related languages or even different spoken varieties of the same language. However, it is still unclear to what extent neural LID models generalize to speech samples with differ…
▽ More
State-of-the-art spoken language identification (LID) systems, which are based on end-to-end deep neural networks, have shown remarkable success not only in discriminating between distant languages but also between closely-related languages or even different spoken varieties of the same language. However, it is still unclear to what extent neural LID models generalize to speech samples with different acoustic conditions due to domain shift. In this paper, we present a set of experiments to investigate the impact of domain mismatch on the performance of neural LID systems for a subset of six Slavic languages across two domains (read speech and radio broadcast) and examine two low-level signal descriptors (spectral and cepstral features) for this task. Our experiments show that (1) out-of-domain speech samples severely hinder the performance of neural LID models, and (2) while both spectral and cepstral features show comparable performance within-domain, spectral features show more robustness under domain mismatch. Moreover, we apply unsupervised domain adaptation to minimize the discrepancy between the two domains in our study. We achieve relative accuracy improvements that range from 9% to 77% depending on the diversity of acoustic conditions in the source domain.
△ Less
Submitted 6 August, 2020; v1 submitted 2 August, 2020;
originally announced August 2020.
-
Thermodynamic properties of Aharonov-Bohm (AB) and magnetic fields with screened Kratzer potential
Authors:
Akpan N. Ikot,
Collins O. Edet,
Precious O. Amadi,
Uduakobong S. Okorie,
G. J. Rampho,
H. B. Abdullah
Abstract:
In this study, the Schrodinger equation (SE) with screened Kratzer potential (SKP) in the presence of external magnetic and AB-flux fields is investigated using the factorization method. The eigenvalue and eigenfunction for the system are obtained in closed form. It is found that the present of the magnetic field partially removes the degeneracy when the screening parameter of the potential was sm…
▽ More
In this study, the Schrodinger equation (SE) with screened Kratzer potential (SKP) in the presence of external magnetic and AB-flux fields is investigated using the factorization method. The eigenvalue and eigenfunction for the system are obtained in closed form. It is found that the present of the magnetic field partially removes the degeneracy when the screening parameter of the potential was small but the addition of the AB field removed the degeneracy faster and better. The magnetization and magnetic susceptibility of the system are evaluated at zero and finite temperatures and other thermodynamic properties of the system are discussed. More so, the presence of the AB-flux field makes the system to exhibit a both a paramagnetic and diamagnetic behavior. A straight forward extension of these results to three dimension shows that the present result is consistent with those obtained in literature.
△ Less
Submitted 14 December, 2019;
originally announced December 2019.
-
Efficient Multiple Incremental Computation for Kernel Ridge Regression with Bayesian Uncertainty Modeling
Authors:
Bo-Wei Chen,
Nik Nailah Binti Abdullah,
Sangoh Park
Abstract:
This study presents an efficient incremental/decremental approach for big streams based on Kernel Ridge Regression (KRR), a frequently used data analysis in cloud centers. To avoid reanalyzing the whole dataset whenever sensors receive new training data, typical incremental KRR used a single-instance mechanism for updating an existing system. However, this inevitably increased redundant computatio…
▽ More
This study presents an efficient incremental/decremental approach for big streams based on Kernel Ridge Regression (KRR), a frequently used data analysis in cloud centers. To avoid reanalyzing the whole dataset whenever sensors receive new training data, typical incremental KRR used a single-instance mechanism for updating an existing system. However, this inevitably increased redundant computational time, not to mention applicability to big streams. To this end, the proposed mechanism supports incremental/decremental processing for both single and multiple samples (i.e., batch processing). A large scale of data can be divided into batches, processed by a machine, without sacrificing the accuracy. Moreover, incremental/decremental analyses in empirical and intrinsic space are also proposed in this study to handle different types of data either with a large number of samples or high feature dimensions, whereas typical methods focused only on one type. At the end of this study, we further the proposed mechanism to statistical Kernelized Bayesian Regression, so that uncertainty modeling with incremental/decremental computation becomes applicable. Experimental results showed that computational time was significantly reduced, better than the original nonincremental design and the typical single incremental method. Furthermore, the accuracy of the proposed method remained the same as the baselines. This implied that the system enhanced efficiency without sacrificing the accuracy. These findings proved that the proposed method was appropriate for variable streaming data analysis, thereby demonstrating the effectiveness of the proposed method.
△ Less
Submitted 8 November, 2017; v1 submitted 1 August, 2016;
originally announced August 2016.
-
Dark Matter Searches at the Large Hadron Collider
Authors:
Siew Yan Hoh,
Jyothsna Komaragiri,
Wan Ahmad Tajuddin Bin Wan Abdullah
Abstract:
Dark Matter is a hypothetical particle proposed to explain the missing matter expected from the cosmological observation. The motivation of Dark Matter is overwhelming however as it is mainly deduced from its gravitational interaction, for it does little to pinpoint what Dark Matter really is. In WIMPs Miracle, weakly interactive massive particle being the Dark Matter candidate is correctly produc…
▽ More
Dark Matter is a hypothetical particle proposed to explain the missing matter expected from the cosmological observation. The motivation of Dark Matter is overwhelming however as it is mainly deduced from its gravitational interaction, for it does little to pinpoint what Dark Matter really is. In WIMPs Miracle, weakly interactive massive particle being the Dark Matter candidate is correctly producing the current thermal relic density at weak scale, implying the possibility of producing and detecting it in Large Hadron Collider. Assuming WIMPs being the maverick particle within collider, it is expected to be pair produced in association with a Standard Model particle. The presence of the WIMPs pair is inferred from the Missing Transverse Energy (MET) which is the vector sum of the imbalance in the transverse momentum plane recoils a Standard Model Particle. The collider is able to produce light mass Dark Matter which the traditional detection fail to detect due to the small momentum transfer involved in the interaction; on the other hand, the traditional detection is robust in detecting a higher Dark matter masses but the collider is su ered from the parton distribution function suppression. Topologically the processes are similar to the scattering processes in the direct detection thus complementary to the traditional Dark Matter detection. The collider searches are strongly motivated as the results are usually translated to the annihilation and scattering rates at more traditional Dark Matter-oriented experiments, thus a concordance approach is adapted. An overview of Dark Matter searches at the Large Hadron Collider will be covered in this paper.
△ Less
Submitted 23 December, 2015;
originally announced December 2015.
-
Energy balancing through cluster head selection using K-Theorem in homogeneous wireless sensor networks
Authors:
Muhammad Imran,
Asfandyar khan,
Azween B. Abdullah
Abstract:
The objective of this paper is to increase life time of homogeneous wireless sensor networks (WSNs) through minimizing long range communication and energy balancing. Sensor nodes are resource constrained particularly with limited energy that is difficult or impossible to replenish. LEACH (Low Energy Adaptive Clustering Hierarchy) is most well-known cluster based architecture for WSN that aims to e…
▽ More
The objective of this paper is to increase life time of homogeneous wireless sensor networks (WSNs) through minimizing long range communication and energy balancing. Sensor nodes are resource constrained particularly with limited energy that is difficult or impossible to replenish. LEACH (Low Energy Adaptive Clustering Hierarchy) is most well-known cluster based architecture for WSN that aims to evenly dissipate energy among all sensor nodes. In cluster based architecture, the role of cluster head is very crucial for the successful operation of WSN because once the cluster head becomes non functional, the whole cluster becomes dysfunctional. We have proposed a modified cluster based WSN architecture by introducing a coordinator node (CN) that is rich in terms of resources. This CN take up the responsibility of transmitting data to the base station over longer distances from cluster heads. We have proposed a cluster head selection algorithm based on K-theorem and other parameters i.e. residual energy, distance to coordinator node, reliability and degree of mobility. The K-theorem is used to select candidate cluster heads based on bunch of sensor nodes in a cluster. We believe that the proposed architecture and algorithm achieves higher energy efficiency through minimizing communication and energy balancing. The proposed architecture is more scalable and proposed algorithm is robust against even/uneven node deployment and node mobility.
△ Less
Submitted 21 December, 2012;
originally announced December 2012.
