While Large Language Models (LLMs) have been increasingly used to generate synthetic data for various downstream tasks, researchers overlook the causal direction in the data synthesis process. A natural causal direction should contain two steps: diverse raw data are generated first, and subsequently annotated for downstream tasks. However, most LLM-based methods adopt an anticausal direction: embedding label information in the prompt to force LLMs to generate targeted data. This reversal raises a critical question: How does the direction of data synthesis impact the quality and utility of the synthetic data? In this work, we empirically study the impact of causal and anticausal data synthesis. To do so, we first design simple yet effective prompting strategies to control the causal direction of LLM-based data synthesis. Using GPT-5 as the data generator, we construct synthetic datasets for three distinct machine learning tasks. We then fine-tune BERT-base and LLaMA-3.2-1B models on these datasets and evaluate them against human-curated benchmarks. Our experiments reveal consistent patterns: (1) models trained on anticausal synthetic data suffer larger performance drops across all tasks and model families --- Accuracy declines range from 13.7%-59.1% for BERT and 4.9%-54.3% for LLaMA, and (2) distributional analysis shows that anticausal synthetic datasets deviate further from human data. Our findings provide practical guidance on how to generate better synthetic data and make good use of it.

Traditional disaster analysis and modelling tools for assessing the severity of a disaster are predictive in nature. Based on the past observational data, these tools prescribe how the current input state (e.g., environmental conditions, situation reports) results in a severity assessment. However, these systems are not meant to be interventional in the causal sense, where the user can modify the current input state to simulate counterfactual ''what-if'' scenarios. In this work, we provide an alternative interventional tool that complements traditional disaster modelling tools by leveraging real-time data sources like satellite imagery, news, and social media. Our tool also helps understand the causal attribution of different factors on the estimated severity, over any given region of interest. In addition, we provide actionable recourses that would enable easier mitigation planning. Our source code is publicly available.

Large Language Models (LLMs) have advanced tremendously in recent years. Large-scale models such as GPT4, GEMINI, Llama, and Claude have opened new frontiers for what is possible for generative models. They’ve been utilized in a plethora of applications, including those that require tedious domain knowledge and subject matter expertise. One such application for the utilization of LLMs proposed in this study is the attribute generation process for both labeled and unlabeled datasets. Data labels, along with the image features/attributes, are crucial for the application of image datasets in causal reasoning tasks. Causal reasoning for image datasets cannot be accomplished without the proper data labels, which serve as the ground truth. However, manual data labeling and attribute/feature generation for image datasets is a tedious and sometimes unfeasible task. Therefore, we propose to use LLMs for automating the attribute generation and image labeling task for images. To address this, we investigate LLMs to determine whether they are capable of generating attributes and how correct these attributes are. Using two different datasets and a combination of prompt-tuning, we highlight mixed results for the ability of ChatGPT 4o to accurately classify the image label and generate attributes for the images based on the prompt, types of images, input methods, and whether fine-tuning is performed. For the AWA dataset, we highlight $100 \%$ accuracy for image classification and attribute generation when prompted with the options for the possible attributes. When the attribute options are not provided the results vary for both datasets. For the MRI scans, the accuracy of the image classification and attributes varied from $40 \%$ to $100 \%$ depending on whether the attribute options were provided and the level of granularity needed.

A Convolutional Neural Network (CNN) relies on intricate pixel correlations to predict objects within images, demonstrating improved performance when identifying objects alongside commonly co-occurring objects. For instance, a helicopter can be more accurately predicted when it appears with an aircraft carrier compared to when it stands alone. Influence on detecting an object due to the presence or absence of another object can be regarded as a form of contextual bias. This phenomenon poses a significant challenge in object detection. Addressing this bias is crucial. Our study explores correlation-aware prediction techniques to disentangle and leverage connections between objects. We propose CART-OD, a novel object detection approach that enhances model performance by adjusting the confidence scores of correlated objects during inference. By strategically intervening during prediction, CART-OD targets instances where commonly co-occurring objects appear separately, effectively mitigating the influence of object correlation. The adjustment in real-time makes the approach computationally efficient and scalable to handle large-scale datasets. Results indicate that CART-OD significantly enhances precision, recall, and mean average precision (mAP) in detecting objects that appear without their most correlated counterparts, underscoring the potential of correlation-aware strategies in object detection tasks.

The evergrowing reliance of humans and society on machine learning methods has raised concerns about their trustworthiness and liability. As a response to these concerns, Socially Responsible Machine Learning (SRML) aims at developing fair, transparent, and robust machine learning algorithms. However, traditional approaches to SRML do not incorporate human perspectives, and therefore are not sufficient to build long-lasting trust between machines and human being. Causality as the key to human intelligence plays a vital role in achieving socially responsible machine learning algorithms which are compatible with human notions. Bridging the gap between traditional SRML and causality, in this tutorial, we aim at providing a holistic overview of SRML through the lens of causality. In particular, we will focus on state-of-the-art techniques on causal socially responsible ML in terms of fairness, interpretability, and robustness. The objectives of this tutorial are as follows: (1) we provide a taxonomy of existing literature on causal socially responsible ML from fairness, interpretability, and robustness perspective; (2) we review the state-of-the-art techniques for each task; and (3) we elucidate open questions and future research directions. We believe this tutorial is beneficial to researchers and practitioners from the areas of data mining, machine learning, and social sciences.

Flood is one of the common natural disasters that can severely affect human life and properties. Early detection, therefore, is of paramount importance to provide help through an emergency response team. Robust flood detection techniques so far have been based on computer vision using images either from cameras, satellite imagery, remote sensing, or radar-based images. However, sound signal-based flood event detection has not been widely explored. In this work, we design an end-to-end architecture for a deep learning-based flood-related sound event detection model. We employ Mel-Spectrogram-based auditory signal analysis and deep learning models for sound event detection (SED). We evaluated four deep learning models under the following two categories: (i) Binary classification Flood/No Flood, vs. Windy vs. Non-Windy, and (ii) Multi-classification for more granular flood and wind events. The experimental results performed in these settings on the datasets collected from real deployment showed an accuracy of around 78%.

Information has a dire role when it comes to decision-making, especially in complex multi-domain operational environments. While it can be assumed that obtaining more information is more beneficial, it can lead towards several challenges. For a decision maker, to collect and assess large quantities of information in order to make an informed decision can be a difficult task. The decision-making task becomes more problematic due to the imperfect nature of information expressed through the uncertainty of information (UoI) concept. This paper proposes implementing multi-criteria decision making (MCDM) within a dynamic multi-agent-based framework called Sentry Agents (SAGE), which allows for the simulation of various decision-making scenarios and diverse decision maker behaviors. We evaluate how MCDM can measure interdependence between criteria, and how that impacts the determination of an ideal course of action in a decision-making task.

Networked observational data presents new opportunities for learning individual causal effects, which plays an indispensable role in decision making. Such data poses the challenge of confounding bias. Previous work presents two desiderata to handle confounding bias. On the treatment group level, we aim to balance the distributions of confounder representations. On the individual level, it is desirable to capture patterns of hidden confounders that predict treatment assignments. Existing methods show the potential of utilizing network information to handle confounding bias, but they only try to satisfy one of the two desiderata. This is because the two desiderata seem to contradict each other. When the two distributions of confounder representations are highly overlapped, then we confront the undiscriminating problem between the treated and the controlled. In this work, we formulate the two desiderata as a minimax game. We propose IGNITE that learns representations of confounders from networked observational data, which is trained by a minimax game to achieve the two desiderata. Experiments verify the efficacy of IGNITE on two datasets under various settings.

Intelligent assistants, such as navigation, recommender, and expert systems, are most helpful in situations where users lack domain knowledge. Despite this, recent research in cognitive psychology has revealed that lower-skilled individuals may maintain a sense of illusory superiority, which might suggest that users with the highest need for advice may be the least likely to defer judgment. Explanation interfaces - a method for persuading users to take a system's advice - are thought by many to be the solution for instilling trust, but do their effects hold for self-assured users? To address this knowledge gap, we conducted a quantitative study (N=529) wherein participants played a binary decision-making game with help from an intelligent assistant. Participants were profiled in terms of both actual (measured) expertise and reported familiarity with the task concept. The presence of explanations, level of automation, and number of errors made by the intelligent assistant were manipulated while observing changes in user acceptance of advice. An analysis of cognitive metrics lead to three findings for research in intelligent assistants: 1) higher reported familiarity with the task simultaneously predicted more reported trust but less adherence, 2) explanations only swayed people who reported very low task familiarity, and 3) showing explanations to people who reported more task familiarity led to automation bias.

Safe handling of classified information is important in private and government organizations. Images are commonly used in military chores, such as, transmitting information of enemy terrain or army's secret base locations. Existing image encryption methods transform an original image into a texture or noise like image. However, an experienced hacker can notice small visual signs identifying an encrypted image and can cyber attack on victim's network. We propose an audio signal processing technique for image encoding. Our contributed approach hides an image by transforming it to an audio signal. Moreover, obtained data is invertible, has better interpretability, and easier to visualize. We provide empirical validation of our approach on a malware image data, well-known for unknown nonlinearity and similarity in layout and texture making it difficult to capture underlying regularities. We set out to understand the impact of malware audio melodies on suitable neighborhood size over which to consider pattern phenomena, if nonlinear methods capture pattern phenomena with increased efficacy, and how they improve clustering and malware identification. In real-life use cases, good clustering practice requires scientific and domain knowledge. As there is no rich domain history of malware audio data, we set out to explore different clustering methods. We developed an information theoretic method for measurement of malware audio melodies and found that nonlinear methods are better in capturing underlying data regularity and give statistically significant improvement in clustering results than linear methods. This is the first application of Manifold learning on malware image-audio encoding in the research literature.

Identification and detection of malware is important in private and government agencies for defending against cyber attack on network security. Current methods for malware analysis achieves promising results using linear methods, such as, PCA. It is well-known that the underlying physical processes behind malware binaries to image conversion are highly nonlinear. Our work set out to understand the impact of malware image resolution on suitable neighborhood size over which to consider pattern phenomena, if nonlinear methods capture pattern phenomena with increased efficacy, and how they improve clustering and identification of malware. In real-life use cases, good clustering practice requires scientific and domain knowledge. As there is no rich domain history of malware image data, we set out to explore different clustering methods with reasoning. Image domains are typically high-dimensional and can have many instantiations. For this reason, statistical approaches to clustering can fail because low probability mass is associated with a single image. Moreover, related images will therefore have subtle differences as measured by their probabilities. We developed an information theoretic method for measurement of malware image resolutions and found that nonlinear methods are better in capturing underlying data regularity and give statistically significant improvement in clustering results than linear methods. This is the first application of nonlinear Manifold learning to malware image analysis in the research literature.

In this paper we consider an application of a WTA dynamics developed from artificial neural networks to the concept of a large array of optoelectronic feedback circuits. The merging of the winner take all (WTA) dynamics with optical implementation could potentially provide improved systems for high resolution image processing. The time required for digital image processing grows dramatically with the increase in image frame resolution, which complicates segmentation, detection, and tracking of objects within an image. This approach could address the challenges found in high resolution image processing. Examples of simulation results based on this approach are presented.