Extending the context length (i.e., the maximum supported sequence length) of LLMs is of paramount significance. To facilitate long context training of LLMs, sequence parallelism has emerged as an essential technique, which scatters each input sequence across multiple devices and necessitates communication to process the sequence. In essence, existing sequence parallelism methods assume homogeneous sequence lengths (i.e., all input sequences are equal in length) and therefore leverages a single, static scattering strategy for all input sequences. However, in reality, the sequence lengths in LLM training corpora exhibit substantial variability, often following a long-tail distribution, which leads to workload heterogeneity. In this paper, we show that employing a single, static strategy results in inefficiency and resource under-utilization, highlighting the need for adaptive approaches to handle the heterogeneous workloads across sequences. To address this, we propose a heterogeneity-adaptive sequence parallelism method. For each training step, our approach captures the variability in sequence lengths and assigns the optimal combination of scattering strategies based on workload characteristics. We model this problem as a linear programming optimization and design an efficient and effective solver to find the optimal solution. Furthermore, we implement our method in a high-performance system that supports adaptive parallelization in distributed LLM training. Experimental results demonstrate that our system outperforms state-of-the-art training frameworks by up to 1.98x.

Driven by recent breakthrough advances in neural representation learning, approximate near-neighbor (ANN) search over vector embeddings has emerged as a critical computational workload. With the introduction of the seminal Hierarchical Navigable Small World (HNSW) algorithm, graph-based indexes have established themseves as the overwhelmingly dominant paradigm for efficient and scalable ANN search. As the name suggests, HNSW searches a layered hierarchical graph to quickly identify neighborhoods of similar points to a given query vector. But is this hierarchy even necessary? A rigorous experimental analysis to answer this question would provide valuable insights into the nature of algorithm design for ANN search and motivate directions for future work in this increasingly crucial domain. To that end, we conduct an extensive benchmarking study covering more large-scale datasets than prior investigations of this question. We ultimately find that a flat graph retains all of the benefits of HNSW on high-dimensional datasets, with latency and recall performance essentially \emph{identical} to the original algorithm but with less memory overhead. Furthermore, we go a step further and study \emph{why} the hierarchy of HNSW provides no benefit in high dimensions, hypothesizing that navigable small world graphs contain a well-connected, frequently traversed ``highway" of hub nodes that maintain the same purported function as the hierarchical layers. We present compelling empirical evidence that the \emph{Hub Highway Hypothesis} holds for real datasets and investigate the mechanisms by which the highway forms. The implications of this hypothesis may also provide future research directions in developing enhancements to graph-based ANN search.

A hallmark of intelligence is the ability to flesh out underspecified situations using "common sense." We propose to extract that common sense from large language models (LLMs), in a form that can feed into probabilistic inference. We focus our investigation on $\textit{guesstimation}$ questions such as "How much are Airbnb listings in Newark, NJ?" Formulating a sensible answer without access to data requires drawing on, and integrating, bits of common knowledge about how $\texttt{Price}$ and $\texttt{Location}$ may relate to other variables, such as $\texttt{Property Type}$. Our framework answers such a question by synthesizing an $\textit{ad hoc}$ probabilistic model. First we prompt an LLM to propose a set of random variables relevant to the question, followed by moment constraints on their joint distribution. We then optimize the joint distribution $p$ within a log-linear family to maximize the overall constraint satisfaction. Our experiments show that LLMs can successfully be prompted to propose reasonable variables, and while the proposed numerical constraints can be noisy, jointly optimizing for their satisfaction reconciles them. When evaluated on probabilistic questions derived from three real-world tabular datasets, we find that our framework performs comparably to a direct prompting baseline in terms of total variation distance from the dataset distribution, and is similarly robust to noise.

We present a novel representation learning method for downstream tasks such as anomaly detection and unsupervised transient classification in high-energy datasets. This approach enabled the discovery of a new fast X-ray transient (FXT) in the Chandra archive, XRT 200515, a needle-in-the-haystack event and the first Chandra FXT of its kind. Recent serendipitous breakthroughs in X-ray astronomy, including FXTs from binary neutron star mergers and an extragalactic planetary transit candidate, highlight the need for systematic transient searches in X-ray archives. We introduce new event file representations, E-t Maps and E-t-dt Cubes, designed to capture both temporal and spectral information, effectively addressing the challenges posed by variable-length event file time series in machine learning applications. Our pipeline extracts low-dimensional, informative features from these representations using principal component analysis or sparse autoencoders, followed by clustering in the embedding space with DBSCAN. New transients are identified within transient-dominant clusters or through nearest-neighbor searches around known transients, producing a catalog of 3,539 candidates (3,427 flares and 112 dips). XRT 200515 exhibits unique temporal and spectral variability, including an intense, hard <10 s initial burst followed by spectral softening in an ~800 s oscillating tail. We interpret XRT 200515 as either the first giant magnetar flare observed at low X-ray energies or the first extragalactic Type I X-ray burst from a faint LMXB in the LMC. Our method extends to datasets from other observatories such as XMM-Newton, Swift-XRT, eROSITA, Einstein Probe, and upcoming missions like AXIS.

Artificial intelligence (AI) is widely used in various fields including healthcare, autonomous vehicles, robotics, traffic monitoring, and agriculture. Many modern AI applications in these fields are multi-tasking in nature (i.e. perform multiple analysis on same data) and are deployed on resource-constrained edge devices requiring the AI models to be efficient across different metrics such as power, frame rate, and size. For these specific use-cases, in this work, we propose a new paradigm of neural network architecture (ILASH) that leverages a layer sharing concept for minimizing power utilization, increasing frame rate, and reducing model size. Additionally, we propose a novel neural network architecture search framework (ILASH-NAS) for efficient construction of these neural network models for a given set of tasks and device constraints. The proposed NAS framework utilizes a data-driven intelligent approach to make the search efficient in terms of energy, time, and CO2 emission. We perform extensive evaluations of the proposed layer shared architecture paradigm (ILASH) and the ILASH-NAS framework using four open-source datasets (UTKFace, MTFL, CelebA, and Taskonomy). We compare ILASH-NAS with AutoKeras and observe significant improvement in terms of both the generated model performance and neural search efficiency with up to 16x less energy utilization, CO2 emission, and training/search time.

Large Language Models (LLMs) have shown impressive performance on various benchmarks, yet their ability to engage in deliberate reasoning remains questionable. We present NYT-Connections, a collection of 358 simple word classification puzzles derived from the New York Times Connections game. This benchmark is designed to penalize quick, intuitive "System 1" thinking, isolating fundamental reasoning skills. We evaluated six recent LLMs, a simple machine learning heuristic, and humans across three configurations: single-attempt, multiple attempts without hints, and multiple attempts with contextual hints. Our findings reveal a significant performance gap: even top-performing LLMs like GPT-4 fall short of human performance by nearly 30%. Notably, advanced prompting techniques such as Chain-of-Thought and Self-Consistency show diminishing returns as task difficulty increases. NYT-Connections uniquely combines linguistic isolation, resistance to intuitive shortcuts, and regular updates to mitigate data leakage, offering a novel tool for assessing LLM reasoning capabilities.

We present ReHub, a novel graph transformer architecture that achieves linear complexity through an efficient reassignment technique between nodes and virtual nodes. Graph transformers have become increasingly important in graph learning for their ability to utilize long-range node communication explicitly, addressing limitations such as oversmoothing and oversquashing found in message-passing graph networks. However, their dense attention mechanism scales quadratically with the number of nodes, limiting their applicability to large-scale graphs. ReHub draws inspiration from the airline industry's hub-and-spoke model, where flights are assigned to optimize operational efficiency. In our approach, graph nodes (spokes) are dynamically reassigned to a fixed number of virtual nodes (hubs) at each model layer. Recent work, Neural Atoms (Li et al., 2024), has demonstrated impressive and consistent improvements over GNN baselines by utilizing such virtual nodes; their findings suggest that the number of hubs strongly influences performance. However, increasing the number of hubs typically raises complexity, requiring a trade-off to maintain linear complexity. Our key insight is that each node only needs to interact with a small subset of hubs to achieve linear complexity, even when the total number of hubs is large. To leverage all hubs without incurring additional computational costs, we propose a simple yet effective adaptive reassignment technique based on hub-hub similarity scores, eliminating the need for expensive node-hub computations. Our experiments on LRGB indicate a consistent improvement in results over the base method, Neural Atoms, while maintaining a linear complexity. Remarkably, our sparse model achieves performance on par with its non-sparse counterpart. Furthermore, ReHub outperforms competitive baselines and consistently ranks among top performers across various benchmarks.

Sign Language Processing (SLP) is an interdisciplinary field comprised of Natural Language Processing (NLP) and Computer Vision. It is focused on the computational understanding, translation, and production of signed languages. Traditional approaches have often been constrained by the use of gloss-based systems that are both language-specific and inadequate for capturing the multidimensional nature of sign language. These limitations have hindered the development of technology capable of processing signed languages effectively. This thesis aims to revolutionize the field of SLP by proposing a simple paradigm that can bridge this existing technological gap. We propose the use of SignWiring, a universal sign language transcription notation system, to serve as an intermediary link between the visual-gestural modality of signed languages and text-based linguistic representations. We contribute foundational libraries and resources to the SLP community, thereby setting the stage for a more in-depth exploration of the tasks of sign language translation and production. These tasks encompass the translation of sign language from video to spoken language text and vice versa. Through empirical evaluations, we establish the efficacy of our transcription method as a pivot for enabling faster, more targeted research, that can lead to more natural and accurate translations across a range of languages. The universal nature of our transcription-based paradigm also paves the way for real-time, multilingual applications in SLP, thereby offering a more inclusive and accessible approach to language technology. This is a significant step toward universal accessibility, enabling a wider reach of AI-driven language technologies to include the deaf and hard-of-hearing community.

Text Simplification is a task that has been minimally explored for low-resource languages. Consequently, there are only a few manually curated datasets. In this paper, we present a human curated sentence-level text simplification dataset for the Sinhala language. Our evaluation dataset contains 1,000 complex sentences and corresponding 3,000 simplified sentences produced by three different human annotators. We model the text simplification task as a zero-shot and zero resource sequence-to-sequence (seq-seq) task on the multilingual language models mT5 and mBART. We exploit auxiliary data from related seq-seq tasks and explore the possibility of using intermediate task transfer learning (ITTL). Our analysis shows that ITTL outperforms the previously proposed zero-resource methods for text simplification. Our findings also highlight the challenges in evaluating text simplification systems, and support the calls for improved metrics for measuring the quality of automated text simplification systems that would suit low-resource languages as well. Our code and data are publicly available: https://github.com/brainsharks-fyp17/Sinhala-Text-Simplification-Dataset-and-Evaluation

The use of Large Language Models (LLMs) has increased significantly recently, with individuals frequently interacting with chatbots to receive answers to a wide range of questions. In an era where information is readily accessible, it is crucial to stimulate and preserve human cognitive abilities and maintain strong reasoning skills. This paper addresses such challenges by promoting the use of hints as an alternative or a supplement to direct answers. We first introduce a manually constructed hint dataset, WIKIHINT, which includes 5,000 hints created for 1,000 questions. We then finetune open-source LLMs such as LLaMA-3.1 for hint generation in answer-aware and answer-agnostic contexts. We assess the effectiveness of the hints with human participants who try to answer questions with and without the aid of hints. Additionally, we introduce a lightweight evaluation method, HINTRANK, to evaluate and rank hints in both answer-aware and answer-agnostic settings. Our findings show that (a) the dataset helps generate more effective hints, (b) including answer information along with questions generally improves hint quality, and (c) encoder-based models perform better than decoder-based models in hint ranking.