Recent jailbreak attempts on Large Language Models (LLMs) have shifted from algorithm-focused to human-like social engineering attacks, with persuasion-based techniques emerging as a particularly effective subset. These attacks evolve rapidly, demonstrate high creativity, and boast superior attack success rates. To combat such threats, we propose a promising approach to enhancing LLM safety by leveraging the underlying geometry of input prompt token embeddings using hypergraphs. This approach allows us to model the differences in information flow between benign and malicious LLM prompts. In our approach, each LLM prompt is represented as a metric hypergraph, forming a compact metric space. We then construct a higher-order metric space over these compact metric hypergraphs using the Gromov-Hausdorff distance as a generalized metric. Within this space of metric hypergraph spaces, our safety filter learns to classify between harmful and benign prompts. Our study presents theoretical guarantees on the classifier's generalization error for novel and unseen LLM input prompts. Extensive empirical evaluations demonstrate that our method significantly outperforms both existing state-of-the-art generic defense mechanisms and naive baselines. Notably, our approach also achieves comparable performance to specialized defenses against algorithm-focused attacks.

Content warning: This paper contains examples of harmful language and content. Recent advances in large language models (LLMs) have made them increasingly vulnerable to jailbreaking attempts, where malicious users manipulate models into generating harmful content. While existing approaches rely on either single-step attacks that trigger immediate safety responses or multi-step methods that inefficiently iterate prompts using other LLMs, we introduce ``Sequence of Context" (SoC) attacks that systematically alter conversational context through strategically crafted context-switching queries (CSQs). We formulate this as a multi-armed bandit (MAB) optimization problem, automatically learning optimal sequences of CSQs that gradually weaken the model's safety boundaries. Our theoretical analysis provides tight bounds on both the expected sequence length until successful jailbreak and the convergence of cumulative rewards. Empirically, our method achieves a 95\% attack success rate, surpassing PAIR by 63.15\%, AutoDAN by 60\%, and ReNeLLM by 50\%. We evaluate our attack across multiple open-source LLMs including Llama and Mistral variants. Our findings highlight critical vulnerabilities in current LLM safeguards and emphasize the need for defenses that consider sequential attack patterns rather than relying solely on static prompt filtering or iterative refinement.

Sindhu Padakandla, Sadbhavana Babar, Rathod Darshan D, Manohar Kaul. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

This paper introduces the Elemental Composite Prototypical Network (ECPN), a novel approach to few-shot learning (FSL) in outdoor 3D point cloud object detection. Such point clouds are inherently non-uniformly packed and show marked intra-class variations due to aberrations in lidar scanning methods. Due to the limited availability of examples in the FSL setting, the intra-class variations serve as a much more formidable challenge to traditional detection algorithms. ECPN employs a novel prototypical learning method that solves the issues mentioned above. We generate and leverage multiple elemental prototypes for each class to capture essential geometric features from limited examples. These elemental prototypes are then combined in a weighted manner to arrive at composite prototypes that score relevant and irrelevant features in the elemental prototypes with respect to the query point cloud scene. Moreover, we introduce a novel feature-similarity-discrimination loss to refine the model's ability to distinguish between relevant objects and their background, significantly improving object detection accuracy in FSL scenarios. Our extensive testing on the nuScenes dataset demonstrates that ECPN significantly outperforms existing baselines, offering a robust solution to the complexities of outdoor few-shot 3D object detection (O-FS3D) and setting a new standard for future research.

Text driven diffusion models have shown remarkable capabilities in editing images. However, when editing 3D scenes, existing works mostly rely on training a NeRF for 3D editing. Recent NeRF editing methods leverages edit operations by deploying 2D diffusion models and project these edits into 3D space. They require strong positional priors alongside text prompt to identify the edit location. These methods are operational on small 3D scenes and are more generalized to particular scene. They require training for each specific edit and cannot be exploited in real-time edits. To address these limitations, we propose a novel method, FreeEdit, to make edits in training free manner using mesh representations as a substitute for NeRF. Training-free methods are now a possibility because of the advances in foundation model's space. We leverage these models to bring a training-free alternative and introduce solutions for insertion, replacement and deletion. We consider insertion, replacement and deletion as basic blocks for performing intricate edits with certain combinations of these operations. Given a text prompt and a 3D scene, our model is capable of identifying what object should be inserted/replaced or deleted and location where edit should be performed. We also introduce a novel algorithm as part of FreeEdit to find the optimal location on grounding object for placement. We evaluate our model by comparing it with baseline models on a wide range of scenes using quantitative and qualitative metrics and showcase the merits of our method with respect to others. Project page: https://vivekmadhavaram.github.io/FreeEdit_page/

In 3D data analysis, point clouds provide detailed geometric insights for applications like computer vision and geospatial analysis. However, their irregularity and diversity make classification challenging, especially in domain generalization, where models must generalize to new data distributions. Our research introduces a novel 3D Domain Generalization (3DDG) method using Unsupervised Part Decomposition (UPD) and Graph Structure Induction (GSI). The UPD module employs spectral clustering and a modified Shannon entropy method to segment point clouds into meaningful parts. The GSI module constructs a graph of these parts' spatial relationships, processed by a Graph Neural Network (GNN) to understand complex geometries. Our approach enhances part-based analysis, improving classification accuracy on the PointDA-10 and GraspNetPC-10 datasets by 1.25% and 2.6%, respectively. These results highlight our advancements in 3D domain generalization, enabling more robust classification models for diverse point cloud data.

Given unstructured text, Large Language Models (LLMs) are adept at answering simple (single-hop) questions.However, as the complexity of the questions increase, the performance of LLMs degrade.We believe this is due to the overhead associated with understanding the complex question followed by filtering and aggregating unstructured information in the raw text.Recent methods try to reduce this burden by integrating structured knowledge triples into the raw text, aiming to provide a structured overview that simplifies information processing.However, this simplistic approach is query-agnostic and the extracted facts are ambiguous as they lack context.To address these drawbacks and to enable LLMs to answer complex (multi-hop) questions with ease, we propose to use a knowledge graph (KG) that is context-aware and is distilled to contain query-relevant information.The use of our compressed distilled KG as input to the LLM results in our method utilizing up to 67% fewer tokens to represent the query relevant information present in the supporting documents, compared to the state-of-the-art (SoTA) method.Our experiments show consistent improvements over the SoTA across several metrics (EM, F1, BERTScore, and Human Eval) on two popular benchmark datasets (HotpotQA and MuSiQue).

Recently, the fundamental problem of unsupervised domain adaptation (UDA) on 3D point clouds has been motivated by a wide variety of applications in robotics, virtual reality, and scene understanding, to name a few. The point cloud data acquisition procedures manifest themselves as significant domain discrepancies and geometric variations among both similar and dissimilar classes. The standard domain adaptation methods developed for images do not directly translate to point cloud data because of their complex geometric nature. To address this challenge, we leverage the idea of multimodality and alignment between distributions. We propose a new UDA architecture for point cloud classification that benefits from multimodal contrastive learning to get better class separation in both domains individually. Further, the use of optimal transport (OT) aims at learning source and target data distributions jointly to reduce the cross-domain shift and provide a better alignment. We conduct a comprehensive empirical study on PointDA-10 and GraspNetPC-10 and show that our method achieves state-of-the-art performance on GraspNetPC-10 (with ≈ 4-12% margin) and best average performance on PointDA-10. Our ablation studies and decision boundary analysis also validate the significance of our contrastive learning module and OT alignment. https://siddharthkatageri.github.io/COT.

Proposing scoring functions to effectively understand, analyze and learn various properties of high dimensional hidden representations of large-scale transformer models like BERT can be a challenging task. In this work, we explore a new direction by studying the topological features of BERT hidden representations using persistent homology (PH). We propose a novel scoring function named “persistence scoring function (PSF)” which: (i) accurately captures the homology of the high-dimensional hidden representations and correlates well with the test set accuracy of a wide range of datasets and outperforms existing scoring metrics, (ii) captures interesting post fine-tuning “per-class” level properties from both qualitative and quantitative viewpoints, (iii) is more stable to perturbations as compared to the baseline functions, which makes it a very robust proxy, and (iv) finally, also serves as a predictor of the attack success rates for a wide category of black-box and white-box adversarial attack methods. Our extensive correlation experiments demonstrate the practical utility of PSF on various NLP tasks relevant to BERT11Code is available at https://github.com/chauhanjatin10/BERTops

In this paper, we present a novel method named RECON, that automatically identifies relations in a sentence (sentential relation extraction) and aligns to a knowledge graph (KG). RECON uses a graph neural network to learn representations of both the sentence as well as facts stored in a KG, improving the overall extraction quality. These facts, including entity attributes (label, alias, description, instance-of) and factual triples, have not been collectively used in the state of the art methods. We evaluate the effect of various forms of representing the KG context on the performance of RECON. The empirical evaluation on two standard relation extraction datasets shows that RECON significantly outperforms all state of the art methods on NYT Freebase and Wikidata datasets.

Several state-of-the-art neural graph embedding methods are based on short random walks (stochastic processes) because of their ease of computation, simplicity in capturing complex local graph properties, scalability, and interpretibility. In this work, we are interested in studying how much a probabilistic bias in this stochastic process affects the quality of the nodes picked by the process. In particular, our biased walk, with a certain probability, favors movement towards nodes whose neighborhoods bear a structural resemblance to the current node's neighborhood. We succinctly capture this neighborhood as a probability measure based on the spectrum of the node's neighborhood subgraph represented as a normalized Laplacian matrix. We propose the use of a paragraph vector model with a novel Wasserstein regularization term. We empirically evaluate our approach against several state-of-the-art node embedding techniques on a wide variety of real-world datasets and demonstrate that our proposed method significantly improves upon existing methods on both link prediction and node classification tasks.

The increased availability of massive point clouds coupled with their utility in a wide variety of applications such as robotics, shape synthesis, and self-driving cars has attracted increased attention from both industry and academia. Recently, deep neural networks operating on labeled point clouds have shown promising results on supervised learning tasks like classification and segmentation. However, supervised learning leads to the cumbersome task of annotating the point clouds. To combat this problem, we propose two novel self-supervised pre-training tasks that encode a hierarchical partitioning of the point clouds using a cover-tree, where point cloud subsets lie within balls of varying radii at each level of the cover-tree. Furthermore, our self-supervised learning network is restricted to pre-train on the support set (comprising of scarce training examples) used to train the downstream network in a few-shot learning (FSL) setting. Finally, the fully-trained self-supervised network's point embeddings are input to the downstream task's network. We present a comprehensive empirical evaluation of our method on both downstream classification and segmentation tasks and show that supervised methods pre-trained with our self-supervised learning method significantly improve the accuracy of state-of-the-art methods. Additionally, our method also outperforms previous unsupervised methods in downstream classification tasks.

The recent proliferation of knowledge graphs (KGs) coupled with incomplete or partial information, in the form of missing relations (links) between entities, has fueled a lot of research on knowledge base completion (also known as relation prediction).Several recent works suggest that convolutional neural network (CNN) based models generate richer and more expressive feature embeddings and hence also perform well on relation prediction.However, we observe that these KG embeddings treat triples independently and thus fail to cover the complex and hidden information that is inherently implicit in the local neighborhood surrounding a triple.To this effect, our paper proposes a novel attention-based feature embedding that captures both entity and relation features in any given entity's neighborhood.Additionally, we also encapsulate relation clusters and multi-hop relations in our model.Our empirical study offers insights into the efficacy of our attention-based model and we show marked performance gains in comparison to state-of-the-art methods on all datasets.

We present an alternate formulation of the partial assignment problem as matching random clique complexes, that are higher-order analogues of random graphs, designed to provide a set of invariants that better detect higher-order structure. The proposed method creates random clique adjacency matrices for each k-skeleton of the random clique complexes and matches them, taking into account each point as the affine combination of its geometric neighborhood. We justify our solution theoretically, by analyzing the runtime and storage complexity of our algorithm along with the asymptotic behavior of the quadratic assignment problem (QAP) that is associated with the underlying random clique adjacency matrices. Experiments on both synthetic and real-world datasets, containing severe occlusions and distortions, provide insight into the accuracy, efficiency, and robustness of our approach. We outperform diverse matching algorithms by a significant margin.

Large-scale graph and machine learning analytics widely employ distributed iterative processing. Typically, these analytics are a part of a comprehensive workflow, which includes data preparation, model building, and model evaluation. General-purpose distributed dataflow frameworks execute all steps of such workflows holistically. This holistic view enables these systems to reason about and automatically optimize the entire pipeline. Here, graph and machine learning analytics are known to incur a long runtime since they require multiple passes over the data until convergence is reached. Thus, fault tolerance and a fast-recovery from any intermittent failure is critical for efficient analysis. In this paper, we propose novel fault-tolerant mechanisms for graph and machine learning analytics that run on distributed dataflow systems. We seek to reduce checkpointing costs and shorten failure recovery times. For graph processing, rather than writing checkpoints that block downstream operators, our mechanism writes checkpoints in an unblocking manner that does not break pipelined tasks. In contrast to the conventional approach for unblocking checkpointing (e.g., that manage checkpoints independently for immutable datasets), we inject the checkpoints of mutable datasets into the iterative dataflow itself. Hence, our mechanism is iteration-aware by design. This simplifies the system architecture and facilitates coordinating checkpoint creation during iterative graph processing. Moreover, we are able to rapidly rebound, via confined recovery, by exploiting the fact that log files exist locally on healthy nodes and managing to avoid a complete recomputation from scratch. In addition, we propose replica recovery for machine learning algorithms, whereby we employ a broadcast variable that enables us to quickly recover without having to introduce any checkpoints. In order to evaluate our fault tolerance strategies, we conduct both a theoretical study and experimental analyses using Apache Flink and discover that they outperform blocking checkpointing and complete recovery.

Real-world graph processing applications often require combining the graph data with tabular data. Moreover, graph processing usually is part of a larger analytics workflow consiting of data preparation, analysis and model building, and model application. General-purpose distributed dataflow frameworks execute all steps of such workflows holistically. This holistic view enables these systems to reason about and automatically optimize the processing. Most big graph processing algorithms are iterative and incur a long runtime, as they require multiple passes over the data until convergence. Thus, fault tolerance and quick recovery from any intermittent failure at any step of the workflow are crucial for effective and efficient analysis. In this work, we propose a novel fault-tolerance mechanism for iterative graph processing on distributed data-flow systems with the objective to reduce the checkpointing cost and failure recovery time. Rather than writing checkpoints that block downstream operators, our mechanism writes checkpoints in an unblocking manner, without breaking pipelined tasks. In contrast to the typical unblocking checkpointing approaches (i.e., managing checkpoints independently for immutable datasets), we inject the checkpoints of mutable datasets into the iterative dataflow itself. Hence, our mechanism is iteration-aware by design. This simplifies the system architecture and facilitates coordinating the checkpoint creation during iterative graph processing. We achieve speedier recovery, i.e., confined recovery, by using the local log files on each node to avoid a complete re-computation from scratch. Our theoretical studies as well as our experimental analysis on Flink give further insight into our fault-tolerance strategies and show that they are more efficient than blocking checkpointing and complete recovery for iterative graph processing on dataflow systems.

The increasing availability of massive and accurate laser data enables the processing of spatial queries on terrains. As shortest-path computation, an integral element of query processing, is inherently expensive on terrains, a key approach to enabling efficient query processing is to reduce the need for exact shortest-path computation in query processing. We develop new lower and upper bounds on terrain shortest distances that are provably tighter than any existing bounds. Unlike existing bounds, the new bounds do not rely on the quality of the triangulation. We show how use of the new bounds speeds up query processing by reducing the need for exact distance computations. Speedups of of nearly an order of magnitude are demonstrated empirically for well-known spatial queries.

Different uses of a road network call for the consideration of different travel costs: in route planning, travel time and distance are typically considered, and green house gas (GHG) emissions are increasingly being considered. Further, travel costs such as travel time and GHG emissions are time-dependent and uncertain. To support such uses, we propose techniques that enable the construction of a multi-cost, time-dependent, uncertain graph (MTUG) model of a road network based on GPS data from vehicles that traversed the road network. Based on the MTUG, we define stochastic skyline routes that consider multiple costs and time-dependent uncertainty, and we propose efficient algorithms to retrieve stochastic skyline routes for a given source-destination pair and a start time. Empirical studies with three road networks in Denmark and a substantial GPS data set offer insight into the design properties of the MTUG and the efficiency of the stochastic skyline routing algorithms.

Terrain data is becoming increasingly popular both in industry and in academia. Many tools have been developed for visualizing terrain data. However, we find that (1) they usually accept very few data formats of terrain data only; (2) they do not support terrain simplification well which, as will be shown, is used heavily for query processing in spatial databases; and (3) they do not provide the surface distance operator which is fundamental for many applications based on terrain data. Motivated by this, we developed a tool called Terrain-Toolkit for terrain data which accepts a comprehensive set of data formats, supports terrain simplification and provides the surface distance operator.

We are witnessing increasing interests in the effective use of road networks. For example, to enable effective vehicle routing, weighted-graph models of transportation networks are used, where the weight of an edge captures some cost associated with traversing the edge, e.g., greenhouse gas (GHG) emissions or travel time. It is a precondition to using a graph model for routing that all edges have weights. Weights that capture travel times and GHG emissions can be extracted from GPS trajectory data collected from the network. However, GPS trajectory data typically lack the coverage needed to assign weights to all edges. This paper formulates and addresses the problem of annotating all edges in a road network with travel cost based weights from a set of trips in the network that cover only a small fraction of the edges, each with an associated ground-truth travel cost. A general framework is proposed to solve the problem. Specifically, the problem is modeled as a regression problem and solved by minimizing a judiciously designed objective function that takes into account the topology of the road network. In particular, the use of weighted PageRank values of edges is explored for assigning appropriate weights to all edges, and the property of directional adjacency of edges is also taken into account to assign weights. Empirical studies with weights capturing travel time and GHG emissions on two road networks (Skagen, Denmark, and North Jutland, Denmark) offer insight into the design properties of the proposed techniques and offer evidence that the techniques are effective.

The use of accurate 3D spatial network models can enable substantial improvements in vehicle routing. Notably, such models enable eco-routing, which reduces the environmental impact of transportation. We propose a novel filtering and lifting framework that augments a standard 2D spatial network model with elevation information extracted from massive aerial laser scan data and thus yields an accurate 3D model. We present a filtering technique that is capable of pruning irrelevant laser scan points in a single pass, but assumes that the 2D network fits in internal memory and that the points are appropriately sorted. We also provide an external-memory filtering technique that makes no such assumptions. During lifting, a triangulated irregular network (TIN) surface is constructed from the remaining points. The 2D network is projected onto the TIN, and a 3D network is constructed by means of interpolation. We report on a large-scale empirical study that offers insight into the accuracy, efficiency, and scalability properties of the framework.

With the increasing availability of terrain data, e.g., from aerial laser scans, the management of such data is attracting increasing attention in both industry and academia. In particular, spatial queries, e.g., k -nearest neighbor and reverse nearest neighbor queries, in Euclidean and spatial network spaces are being extended to terrains. Such queries all rely on an important operation, that of finding shortest surface distances. However, shortest surface distance computation is very time consuming. We propose techniques that enable efficient computation of lower and upper bounds of the shortest surface distance, which enable faster query processing by eliminating expensive distance computations. Empirical studies show that our bounds are much tighter than the best-known bounds in many cases and that they enable speedups of up to 43 times for some well-known spatial queries.

The reduction of greenhouse gas (GHG) emissions from transportation is essential for achieving politically agreed upon emissions reduction targets that aim to combat global climate change. So-called eco-routing and eco-driving are able to substantially reduce GHG emissions caused by vehicular transportation. To enable these, it is necessary to be able to reliably quantify the emissions of vehicles as they travel in a spatial network. Thus, a number of models have been proposed that aim to quantify the emissions of a vehicle based on GPS data from the vehicle and a 3D model of the spatial network the vehicle travels in. We develop an evaluation framework, called EcoMark, for such environmental impact models. In addition, we survey all eleven state-of-the-art impact models known to us. To gain insight into the capabilities of the models and to understand the effectiveness of the EcoMark, we apply the framework to all models.

Emerging trends in urban mobility have accelerated the need for effective traffic prediction and management systems. The present paper proposes a novel approach to using continuously streaming moving object trajectories for traffic prediction and management. The approach continuously performs three functions for streams of moving object positions in road networks: 1) management of current evolving trajectories, 2) incremental mining of closed frequent routes, and 3) prediction of near-future locations and densities based on 1) and 2). The approach is empirically evaluated on a large real-world data set of moving object trajectories, originating from a fleet of taxis, illustrating that detailed closed frequent routes can be efficiently discovered and used for prediction.