AI-generated code has become an integral part of the mainstream developer workflow today. However, in community-driven platforms like Stack Overflow (SO), where trust, authorship, and credibility are important, it can lead to serious complications. While recent studies have focused on detecting AI-generated code, they have mostly worked with long code samples from repositories and assignments. In contrast, code snippets on SO are often small and context-specific, and thus may prove more challenging for detection. Moreover, another aspect overlooked in prior studies concerns recognizing adversarially prompted AI code deliberately crafted to resemble human-written code. To address these limitations, we have first introduced a large-scale dataset comprising 3500 pairs of SO and ChatGPT answers, along with a curated set of 4500 adversarially prompted AI responses. Next, we evaluate existing code language models over this newly curated dataset. Our evaluation shows that existing models perform well on standard AI answers but fail to detect adversarial ones. Finally, to improve detection, we propose an ensemble approach combining stylometric features of code along with the code embeddings. Our approach shows consistent improvements across multiple models and improves resistance to adversarial prompted code. Our overall findings open promising directions for future research into understanding the nuances of AI code detection with adversarial prompting and code stylometry.

The field of varying feature space in online learning settings, also known as haphazard inputs, is very prominent nowadays due to its applicability in various fields. However, the current solutions to haphazard inputs are model-dependent and cannot benefit from the existing advanced deep-learning methods, which necessitate inputs of fixed dimensions. Therefore, we propose to transform the varying feature space in an online learning setting to a fixed-dimension image representation on the fly. This simple yet novel approach is model-agnostic, allowing any vision-based models to be applicable for haphazard inputs, as demonstrated using ResNet and ViT. The image representation handles the inconsistent input data seamlessly, making our proposed approach scalable and robust. We show the efficacy of our method on four publicly available datasets. The code is available at https://github.com/Rohit102497/HaphazardInputsAsImages.

Stack Overflow is one of the prominent resources in the software community where developers seek technical assistance. Owing to its popularity, the platform witnesses questions of varying difficulty levels. Since not all practitioners are equally adept at addressing these queries, many queries remain unanswered or experience delays in receiving responses. However, the current framework of the site primarily categorizes the questions based on programming languages and specific topics, not considering the complexity level of the post. To bridge this gap, we introduce DATSO, a Difficulty Assessment Tool for Stack Overflow Questions. DATSO is a Chrome extension that assigns “difficulty level tags” to Stack Overflow posts by leveraging the textual and context-dependent features of the questions. Unlike previous works, which face issues like data imbalance and cold start problems, DATSO overcomes these limitations, achieving 72.3% accuracy on a benchmark dataset of Java questions, surpassing existing baseline works by 7%. The video demonstration and code repository can be found at https://youtu.be/VSu2Q0zb-X0 and https://github.com/krish10924/Difficulty-tag-Generator.

Accurate classification of microscopy images is critical for the analysis of biological samples. The availability of large-scale labeled datasets has contributed to recent progress in training large, deep classification models in the medical imaging domain, but methods that cater to a variety of microscopy modalities across a range of biological samples and length scales are scarce. A key reason is that curating labeled data for microscopy images is costly and needs tedious and timeconsuming effort of AI and domain experts. We propose a novel few-shot learning technique, specifically “Blend & Predict” that uses small labeled datasets for training and infers unlabeled datasets. We evaluated the performance and generalizability of our approach using three medical image datasets, each with a different microscopy modality and addressing a different biomedical question on different samples. We achieved results comparable to state-of-the-art models like GoogleNet, VGG16, RestNet50 that used large datasets for training.

Detecting and analyzing nanoscale motion patterns of vesicles, smaller than the microscope resolution (~250 nm), inside living biological cells is a challenging problem. State-of-the-art CV approaches based on detection, tracking, optical flow or deep learning perform poorly for this problem. We propose an integrative approach, built upon physics based simulations, nanoscopy algorithms, and shallow residual attention network to make it possible for the first time to analysis sub-resolution motion patterns in vesicles that may also be of sub-resolution diameter. Our results show state-of-the-art performance, 89% validation accuracy on simulated dataset and 82% testing accuracy on an experimental dataset of living heart muscle cells imaged under three different pathological conditions. We demonstrate automated analysis of the motion states and changed in them for over 9000 vesicles. Such analysis will enable large scale biological studies of vesicle transport and interaction in living cells in the future.

Subspace techniques have been introduced in the framework of contrast source (CS) extended born (CSEB) model, for improving its reconstruction capabilities. Two techniques are demonstrated. First, a scheme for generating a good initial guess of the scatterer profile is shown. Second, subspace-based optimization method is used for optimization. Using the suggested techniques, CSEB model can be applied for solving inverse electromagnetic scattering problem with an extended range of application with respect to previous contributions, particularly for very high contrast lossy scatterers.

This paper proposes an improved subspace-based optimization method (SOM) by using a new construction method for the ambiguous part of the induced current. The new current construction method reduces not only the computational complexity of the current construction in every iteration of the optimization but also the computational complexity of the singular-value decomposition of the mapping from the induced current to scattered fields. Thus, the improved SOM is able to deal with the 3-D inverse-scattering problems. Numerical tests validate the algorithm.

Electromagnetic inverse techniques are non-destructive techniques to investigate an unknown region. These techniques use the principle of scattering to determine the number of objects present in the domain, their properties and shapes. However, the scattered field is non-linear function of the objects’ parameters. Direct search methods prove beneficial in solving such problems. In this paper, we study a two-dimensional domain having dielectric elliptic cylinders of infinite length. We try to estimate the location, contour and relative permittivity of the each of the cylinders. The previous works have majorly contributed to optimization of shapes of cylinders made of perfect electric conductor. Here, we investigate cases of domain having single dielectric elliptic cylinder in different orientations and in noisy/noise-free scenarios. We also present results for a noise-free domain containing two dielectric elliptic cylinders. We use Multiple Signal Classification algorithm to find the exact number of cylinders in the domain and their locations. Then, Differential Evolution is used to estimate the relative permittivities and contours of the cylinders.