Abstract

The rapid expansion of the development of hydropower in geologically and seismically active areas has numerous technological risks involved. Tunnels in rock masses have different stress and deformation responses than other engineering materials due to the existence of weak planes. This paper describes the use of numerical modeling approaches to assess tunnel deformation and tunnel stability under the earthquake loading for an 8.8 m horseshoe-shaped power tunnel excavated in Vishnugarh Pipalkoti, Uttaranchal, India. In this analysis, two different approaches are used to perform pseudo-static analyzes to understand the impacts of seismicity on tunnels. The rock mass is modeled as continuum in the first approach, and the tunnel is treated as a continuum equivalent with the inclusion of critical joint sets in the

Abstract

In this paper, we describe our submissions for the Similar Language Translation Shared Task 2021. We built 3 systems in each direction for the Tamil ⇐⇒ Telugu language pair. This paper outlines experiments with various tokenization schemes to train statistical models. We also report the configuration of the submitted systems and results produced by them.

Abstract

This paper presents the application of computer vision and machine learning to autonomous approach and landing and taxiing for an air vehicle. Recently, there has been growing interest in developing unmanned aircraft systems (UAS). We present a system and method that uses pattern recognition which aids the landing of a UAS and enhances the human-crewed air vehicle landing. Auto-landing systems based on the Instrument Landing System (ILS) have already proven their importance through decades. The auto-land systems work in conjunction with a radio altimeter, ILS, MLS, or GNSS. Closer to the runway, both under VFR and IFR, pilots are expected to rely on visual references for landing. Modern systems like HUD or CVS allow a trained pilot to manually fly the aircraft using guidance cues from the flight guidance system.Notwithstanding the type of landing and instruments used, typically, Pilots are expected to have the runway threshold markings, aiming point, displacement arrows, and touch down markings/lights insight before Minimum Decision Altitude (MDA). Imaging sensors are the essential standard equipment in crewed and crewless aerial vehicles that are widely used during the landing maneuver. In this method, a dataset of visual objects from satellite images is subjected to pattern recognition training. This trained system learns and then identifies and locates important visual references from imaging sensors and could help in landing and taxiing.

Abstract

In this work, NiMg@OH/reduced graphene oxide (rGO) nanocomposites are synthesised through microwave-assisted synthetic methodology using various reducing agents. The structural features and surface morphology of synthesised materials are fully examined with different spectroscopy and microscopy techniques such as X-ray diffraction (XRD), scanning electron microscopy (SEM), UV-visible spectroscopy (UV-Vis), Raman and Fourier Transform Infrared (FT-IR) spectroscopy. Further, this material was subjected to cyclic voltammetry (CV) and electrochemical impedance spectroscopy (EIS) techniques for the systematic study of the properties and impact of reducing agents at various percentages of rGO on the specific capacitance of synthesised materials. The presence of rGO gradually improves the charge transfer process at the electrode/electrolyte interface. Among these nanocomposite materials, maximum specific capacitance of 650.31 F/g was observed for NiMg@OH-1.5 wt% rGO synthesised using NH4OH reducing agent in 6 M KOH electrolyte. Hence, the result signifies these composite materials are a promising

Abstract

Automatic identification of emotions from a given music track would be of significant importance in the area of Music Information Retrieval. Several studies have found lyrics an important feature in the task of emotion classification [1]. Using lyrics, till date, a transformer-model which is a Natural-Language-Processing (NLP) based Deep-Learning model has achieved state-of-the-art results in identifying music emotion as represented in the Valence-Arousal (VA) space [2]. While transformer models rely on parsing the syntactic structure of sequential input data, we assume that semantic relations could be efficiently learned from these representations using novel capsule networks [3]. The “capsules” are akin to neuronal populations breaking down information into parts that mimics human processing of information by chunking for efficient encoding and evaluation of the stimulus. We extend the same logic further to say that capsule networks should be able to look at the semantic representation of each word in the lyrics of a song and should be able to learn the relationship between these words and the emotion that they represent as a group

Abstract

Recent studies utilizing online music streaming data have found genre preferences to be associated with risk for depression (Surana et al., 2020). However, the mentioned study does not address gender differences among participants. Gender differences have been reported in depression in terms of the prevalence, symptomatology, and emotion regulation strategies used (Garnefski et al., 2004). Also, there exists a gender-specific tendency to choose certain kinds of music where females tend to prefer ‘softer’ styles like mainstream pop, while males are inclined towards ‘heavier’ styles like hard rock (North and Hargreaves, 2008). Martin et al.(1993) investigated the relationship between adolescents’ music preferences and risk factors for suicide, which included depression. Females preferring rock/metal music were found to have higher scores for depression as compared to males preferring rock/metal. Miranda and Claes (2009) found metal music-listening to be associated with higher depression levels in adolescent females, but only if they affiliate with peers who are more depressed. However, the above studies relied on self-reported music preferences. The current study explores these associations in more naturalistic settings using online music consumption.

Abstract

Personality is known to influence emotional experiences including those induced by music listening. Previous research has largely relied on self-report and models with limited terms, which may not adequately capture naturalistic behavior and experience. Data from the platform Last.fm allows for the analysis of listening behaviors using both musical features and semantic tags provided by listeners, for a more naturalistic exploration of personality, music and emotion.

Abstract

Studies have emerged characterizing naturally occurring music listening behaviour of individuals at-risk for depression via the online music streaming platform Last.fm (Surana et al. 2020a; Surana et al. 2020b), albeit limited to social tags and acoustic features. No studies have looked at artist preferences and predictive capacity using this information associated with risk for depression.

Abstract

Large pretrained models have seen enormous success in extractive summarization tasks. In this work, we investigate the influence of pretraining on a BERT-based extractive summarization system for scientific documents. We derive significant performance improvements using an intermediate pretraining step that leverages existing summarization datasets and report state-of-the-art results on a recently released scientific summarization dataset, SciTLDR. We systematically analyze the intermediate pretraining step by varying the size and domain of the pretraining corpus, changing the length of the input sequence in the target task and varying target tasks. We also investigate how intermediate pretraining interacts with contextualized word embeddings trained on different domains.

Abstract

Principal component analysis(PCA) is one of the most popular linear dimensionality reduction techniques in machine learning. In this paper, we try to present a method for performing PCA on encrypted data using a homomorphic encryption scheme. In a client-server model where the server performs computations on the encrypted data, it (server) does not require to perform any matrix operations like multiplication, inversion, etc. on the encrypted data. This reduces the number of computations significantly since matrix operations on encrypted data are very computationally expensive. For our purpose, we used the CKKS homomorphic encryption scheme since it is most suitable for machine learning tasks allowing approximate computations on real numbers. We also present the experimental results of our proposed Homomorphic PCA(HPCA) algorithm on a few datasets. We measure the R2 score on the reconstructed data and use it as an evaluation metric for our HPCA algorithm. Keywords: Homomorphic Encryption · Principal Component Analysis(PCA) · CKKS scheme · Goldschmidt’s Algorithm