Priyanshu Gupta, Shashank Kirtania, Ananya Singha, Sumit Gulwani, Arjun Radhakrishna, Gustavo Soares, Sherry Shi. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Seismic surface wave dispersion analysis provides insights into the subsurface characteristics of the Earth. High-resolution dispersion images provide a reliable estimate of the group velocity dispersion curve, which is obtained by picking the maxima at central frequencies in the dispersion image and, therefore, results in a robust tomographic structure of the Earth upon inversion. Uncertainties in picking maxima at central frequencies need to be minimized for accurate tomography of the Earth. In this article, a novel technique to estimate super-resolution dispersion images is proposed. Superlet transform (SLT), which is an extension of the continuous wavelet transform, is used to estimate the surface wave dispersion images. Superlets (SL) are the set of wavelets with increasing wavelet cycles along with the order of wavelets, which improves frequency resolution at higher frequencies and reduces redundancy in representing time-frequency superlet coefficients. The performance of the method proposed in this article is compared with the conventional method based on the continuous wavelet transform (CWT) to estimate group velocity dispersion curves of surface waves. The results using synthetic test data show that the group velocities in the dispersion image can be more effectively measured using the proposed method than using the CWT-based approach. This is confirmed by quantitatively measuring the root mean square error (RMSE) between the estimated and theoretical group velocity dispersion curve. The proposed method has also been evaluated using real ambient noise data from the USArray (USA).

Developers spend a significant amount of time in editing code for a variety of reasons such as bug fixing or adding new features. Designing effective methods to predict code edits has been an active yet challenging area of research due to the diversity of code edits and the difficulty of capturing the developer intent. In this work, we address these challenges by endowing pre-trained large language models (LLMs) with the knowledge of relevant prior associated edits, which we call the Grace (Generation conditioned on Associated Code Edits) method. The generative capability of the LLMs helps address the diversity in code changes and conditioning code generation on prior edits helps capture the latent developer intent. We evaluate two well-known LLMs, codex and CodeT5, in zero-shot and fine-tuning settings respectively. In our experiments with two datasets, Grace boosts the performance of the LLMs significantly, enabling them to generate 29% and 54% more correctly edited code in top-1 suggestions relative to the current state-of-the-art symbolic and neural approaches, respectively.

Surface wave tomography is performed by first estimating surface wave dispersion curves and then inverting them. The objective of this paper is the efficient and reliable estimation of the group velocity dispersion curves of surface waves. In this paper, group velocity dispersion curves of surface waves recorded at a single station with known location of seismic source or between a pair of stations are estimated using continuous wavelet transform (CWT). The advantage of CWT is its multi-resolution property and flexible choice of analyzing function. In the proposed method, the number of CWT filters used are almost half of the total number of filters used in the conventional frequency-time analysis (FTAN) method. The CWT coefficients of the seismogram are the functions of time and scale (analogues to wave period). The arrival time to reach the peak of the envelope function of CWT coefficients is estimated to calculate group velocity of the surface waves. Group velocity dispersion curve of the surface waves is the plot of change of group velocity with the period. Synthetic test data and ambient noise recordings of MesoAmerica Seismic Experiment (MASE) stations are acquired to investigate the performance of the proposed method. It is observed that the proposed method effectively retrieves the weaker surface waves and results the broader band group-velocity dispersion curve when compared to the conventional FTAN method. The proposed method is also found to be computationally efficient.

Integrated Development Environments (IDEs) provide tool support to automate many source code editing tasks. Traditionally, IDEs use only the spatial context, i.e., the location where the developer is editing, to generate candidate edit recommendations. However, spatial context alone is often not sufficient to confidently predict the developer’s next edit, and thus IDEs generate many suggestions at a location. Therefore, IDEs generally do not actively offer suggestions and instead, the developer is usually required to click on a specific icon or menu and then select from a large list of potential suggestions. As a consequence, developers often miss the opportunity to use the tool support because they are not aware it exists or forget to use it. To better understand common patterns in developer behavior and produce better edit recommendations, we can additionally use the temporal context, i.e., the edits that a developer was recently performing. To enable edit recommendations based on temporal context, we present Overwatch, a novel technique for learning edit sequence patterns from traces of developers’ edits performed in an IDE. Our experiments show that Overwatch has 78% precision and that Overwatch not only completed edits when developers missed the opportunity to use the IDE tool support but also predicted new edits that have no tool support in the IDE.

Extreme Classification (XC) refers to supervised learning where each training/test instance is labeled with small subset of relevant labels that are chosen from a large set of possible target labels. The framework of XC has been widely employed in web applications such as automatic labeling of web-encyclopedia, prediction of related searches, and recommendation systems.

Human iris contains rich textural information which serves as the key information for biometric identifications. It is very unique and one of the most accurate biometric modalities. However, spoofing techniques can be used to obfuscate or impersonate identities and increase the risk of false acceptance or false rejection. This paper revisits iris recognition with spoofing attacks and analyzes their effect on the recognition performance. Specifically, print attack with contact lens variations is used as the spoofing mechanism. It is observed that print attack and contact lens, individually and in conjunction, can significantly change the inter-personal and intra-personal distributions and thereby increase the possibility to deceive the iris recognition systems. The paper also presents the IIITD iris spoofing database, which contains over 4800 iris images pertaining to over 100 individuals with variations due to contact lens, sensor, and print attack. Finally, the paper also shows that cost effective descriptor approaches may help in counter-measuring spooking attacks.