arXiv subjects

Viktor Prasanna

Publications and source records attributed to Viktor Prasanna.

At least 37 records · Page 2Linked to original sources

arXiv2026

Deep Meta Coordination Graphs for Multi-agent Reinforcement Learning

This paper presents deep meta coordination graphs (DMCG) for learning cooperative policies in multi-agent reinforcement learning (MARL). Coordination graph formulations encode local interactions and accordingly factorize the joint value function of all agents to improve efficiency in MARL. Through DMCG, we dynamically compose what we refer to as \textit{meta coordination graphs}, to learn a more expressive representation of agent interactions and use them to integrate agent information through graph convolutional networks. The goal is to enable an evolving coordination graph to guide effective coordination in cooperative MARL tasks. The graphs are jointly optimized with agents' value functions to learn to implicitly reason about joint actions, facilitating the end-to-end learning of interaction representations and coordinated policies. We demonstrate that DMCG consistently achieves state-of-the-art coordination performance and sample efficiency on challenging cooperative tasks, outperforming several prior graph-based and non-graph-based MARL baselines. Through several ablations, we also isolate the impact of individual components in DMCG, showing that the observed improvements are due to the meaningful design choices in this approach. We also include an analysis of its computational complexity to discuss its practicality in real-world applications. All codes can be found here: {\color{blue}{https://github.com/Nikunj-Gupta/dmcg-marl}.

cs.LG↗

arXiv2026

ConsensusDrop: Fusing Visual and Cross-Modal Saliency for Efficient Vision Language Models

Vision-Language Models (VLMs) are expensive because the LLM processes hundreds of largely redundant visual tokens. Existing token reduction methods typically exploit \textit{either} vision-encoder saliency (broad but query-agnostic) \textit{or} LLM cross-attention (query-aware but sparse and costly). We show that neither signal alone is sufficient: fusing them consistently improves performance compared to unimodal visual token selection (ranking). However, making such fusion practical is non-trivial: cross-modal saliency is usually only available \emph{inside} the LLM (too late for efficient pre-LLM pruning), and the two signals are inherently asymmetric, so naive fusion underutilizes their complementary strengths. We propose \textbf{ConsensusDrop}, a training-free framework that derives a \emph{consensus} ranking by reconciling vision encoder saliency with query-aware cross-attention, retaining the most informative tokens while compressing the remainder via encoder-guided token merging. Across LLaVA-1.5/NeXT, Video-LLaVA, and other open-source VLMs, ConsensusDrop consistently outperforms prior pruning methods under identical token budgets and delivers a stronger accuracy-efficiency Pareto frontier -- preserving near-baseline accuracy even at aggressive token reductions while reducing TTFT and KV cache footprint. Our code will be open-sourced.

cs.CV↗

arXiv2026

System-Level Performance Modeling of Photonic In-Memory Computing

Photonic in-memory computing is a high-speed, low-energy alternative to traditional transistor-based digital computing that utilizes high photonic operating frequencies and bandwidths. In this work, we develop a comprehensive system-level performance model for photonic in-memory computing, capturing the effects of key latency sources such as external memory access and opto-electronic conversion. We perform algorithm-to-hardware mapping across a range of workloads, including the Sod shock tube problem, Matricized Tensor Times Khatri-Rao Product (MTTKRP), and the Vlasov-Maxwell equation, to evaluate how the latencies impact real-world high-performance computing workloads. Our performance model shows that, while accounting for system overheads, a compact 1x256 bit single-wavelength photonic SRAM array, fabricated using the standard silicon photonics process by GlobalFoundries, sustains up to 1.5 TOPS, 0.9 TOPS, and 1.3 TOPS on the Sod shock tube problem, MTTKRP, and the Vlasov-Maxwell equation with an average energy efficiency of 2.5 TOPS/W.

cs.DC↗

arXiv2026

Primitive-Driven Acceleration of Hyperdimensional Computing for Real-Time Image Classification

Hyperdimensional Computing (HDC) represents data using extremely high-dimensional, low-precision vectors, termed hypervectors (HVs), and performs learning and inference through lightweight, noise-tolerant operations. However, the high dimensionality, sparsity, and repeated data movement involved in HDC make these computations difficult to accelerate efficiently on conventional processors. As a result, executing core HDC operations: binding, permutation, bundling, and similarity search: on CPUs or GPUs often leads to suboptimal utilization, memory bottlenecks, and limits on real-time performance. In this paper, our contributions are two-fold. First, we develop an image-encoding algorithm that, similar in spirit to convolutional neural networks, maps local image patches to hypervectors enriched with spatial information. These patch-level hypervectors are then merged into a global representation using the fundamental HDC operations, enabling spatially sensitive and robust image encoding. This encoder achieves 95.67% accuracy on MNIST and 85.14% on Fashion-MNIST, outperforming prior HDC-based image encoders. Second, we design an end-to-end accelerator that implements these compute operations on an FPGA through a pipelined architecture that exploits parallelism both across the hypervector dimensionality and across the set of image patches. Our Alveo U280 implementation delivers 0.09ms inference latency, achieving up to 1300x and 60x speedup over state-of-the-art CPU and GPU baselines, respectively.

cs.AR↗

arXiv2026

SPARC-RAG: Adaptive Sequential-Parallel Scaling with Context Management for Retrieval-Augmented Generation

Retrieval-Augmented Generation (RAG) grounds large language model outputs in external evidence, but remains challenged on multi-hop question answering that requires long reasoning. Recent works scale RAG at inference time along two complementary dimensions: sequential depth for iterative refinement and parallel width for coverage expansion. However, naive scaling causes context contamination and scaling inefficiency, leading to diminishing or negative returns despite increased computation. To address these limitations, we propose SPARC-RAG, a multi-agent framework that coordinates sequential and parallel inference-time scaling under a unified context management mechanism. SPARC-RAG employs specialized agents that maintain a shared global context and provide explicit control over the scaling process. It generates targeted, complementary sub-queries for each branch to enable diverse parallel exploration, and explicitly regulates exiting decisions based on answer correctness and evidence grounding. To optimize scaling behavior, we further introduce a lightweight fine-tuning method with process-level verifiable preferences, which improves the efficiency of sequential scaling and effectiveness of parallel scaling. Across single- and multi-hop QA benchmarks, SPARC-RAG consistently outperforms previous RAG baselines, yielding an average +6.2 F1 improvement under lower inference cost.

cs.IR↗

arXiv2025

Enabling Long FFT Convolutions on Memory-Constrained FPGAs via Chunking

The need for long-context reasoning has led to alternative neural network architectures besides Transformers and self-attention, a popular model being Hyena, which employs causal 1D-convolutions implemented with FFTs. Long convolutions enable efficient global context mixing, but requirements for intermediate results exceed the 2-3 MB Block RAM capacity of FPGAs. We present a chunked FFT convolution approach enabling 450K length sequence by 450K length filter convolutions on an Alveo U200 FPGA with 2.8 MB BRAM through chunking and overlap-add reconstruction. We find that throughput scales proportionally with chunk size while degrading minimally by 7% for our longest sequences, demonstrating that careful memory management enables deployment of long-context primitives on edge FPGAs without sacrificing performance.

cs.LG↗

arXiv2025

Context-Driven Performance Modeling for Causal Inference Operators on Neural Processing Units

The proliferation of large language models has driven demand for long-context inference on resource-constrained edge platforms. However, deploying these models on Neural Processing Units (NPUs) presents significant challenges due to architectural mismatch: the quadratic complexity of standard attention conflicts with NPU memory and compute patterns. This paper presents a comprehensive performance analysis of causal inference operators on a modern NPU, benchmarking quadratic attention against sub-quadratic alternatives including structured state-space models and causal convolutions. Our analysis reveals a spectrum of critical bottlenecks: quadratic attention becomes severely memory-bound with catastrophic cache inefficiency, while sub-quadratic variants span from compute-bound on programmable vector cores to memory-bound by data movement. These findings provide essential insights for co-designing hardware-aware models and optimization strategies to enable efficient long-context inference on edge platforms.

cs.DC↗

arXiv2025

LYNX: Learning Dynamic Exits for Confidence-Controlled Reasoning

Large reasoning models achieve strong performance on complex tasks by generating extended chains of thought, but they often "overthink": continuing to reason long after they have enough information to answer correctly. This wastes inference-time compute and can hurt accuracy. Existing attempts to stop early either manipulate decoding with extra sampling and heuristics, rely on auxiliary verifier models, or operate only as post-hoc analysis pipelines without formal guarantees. We introduce LYNX, an online early-exit mechanism that turns a model's own hidden-state awareness into confidence-controlled stopping decisions. LYNX attaches exit decisions to naturally occurring reasoning cues (e.g., "hmm", "wait") during generation, trains a lightweight probe on hidden states at those cue tokens using supervision from forced exits, and wraps the resulting scores in split conformal prediction to obtain distribution-free control over premature exits. Crucially, we train and calibrate this probe once on a generic mathematical corpus and reuse it unchanged across benchmarks, decoding temperatures, and even non-mathematical tasks. Across three model families spanning 1.5B to 32B parameters, a single mathematically trained probe per base model yields strong accuracy--efficiency tradeoffs. On GSM8K, LYNX matches or improves baseline accuracy while reducing tokens by 40--65\%; on MATH-500 it improves accuracy by up to 12 points with roughly 35--60\% fewer tokens; on AIME 2024 it recovers baseline accuracy with more than 50\% token savings; and on CommonsenseQA, a non-math benchmark, it transfers zero-shot with modest accuracy gains and up to 70\% fewer tokens. Compared to state-of-the-art early-exit methods, LYNX offers competitive or superior Pareto frontiers while remaining fully online, requiring no proxy models at inference, and providing explicit, user-tunable confidence guarantees.

cs.CL↗

arXiv2025

Mixture of Scope Experts at Test: Generalizing Deeper Graph Neural Networks with Shallow Variants

Heterophilous graphs, where dissimilar nodes tend to connect, pose a challenge for graph neural networks (GNNs). Increasing the GNN depth can expand the scope (i.e., receptive field), potentially finding homophily from the higher-order neighborhoods. However, GNNs suffer from performance degradation as depth increases. Despite having better expressivity, state-of-the-art deeper GNNs achieve only marginal improvements compared to their shallow variants. Through theoretical and empirical analysis, we systematically demonstrate a shift in GNN generalization preferences across nodes with different homophily levels as depth increases. This creates a disparity in generalization patterns between GNN models with varying depth. Based on these findings, we propose to improve deeper GNN generalization while maintaining high expressivity by Mixture of scope experts at test (Moscat). Experimental results show that Moscat works flexibly with various GNNs across a wide range of datasets while significantly improving accuracy. Our code is available at (https://github.com/Hydrapse/moscat).

cs.LG↗

arXiv2025

TIGER-MARL: Enhancing Multi-Agent Reinforcement Learning with Temporal Information through Graph-based Embeddings and Representations

In this paper, we propose capturing and utilizing \textit{Temporal Information through Graph-based Embeddings and Representations} or \textbf{TIGER} to enhance multi-agent reinforcement learning (MARL). We explicitly model how inter-agent coordination structures evolve over time. While most MARL approaches rely on static or per-step relational graphs, they overlook the temporal evolution of interactions that naturally arise as agents adapt, move, or reorganize cooperation strategies. Capturing such evolving dependencies is key to achieving robust and adaptive coordination. To this end, TIGER constructs dynamic temporal graphs of MARL agents, connecting their current and historical interactions. It then employs a temporal attention-based encoder to aggregate information across these structural and temporal neighborhoods, yielding time-aware agent embeddings that guide cooperative policy learning. Through extensive experiments on two coordination-intensive benchmarks, we show that TIGER consistently outperforms diverse value-decomposition and graph-based MARL baselines in task performance and sample efficiency. Furthermore, we conduct comprehensive ablation studies to isolate the impact of key design parameters in TIGER, revealing how structural and temporal factors can jointly shape effective policy learning in MARL. All codes can be found here: https://github.com/Nikunj-Gupta/tiger-marl.

cs.LG↗

arXiv2025

Training Diverse Graph Experts for Ensembles: A Systematic Empirical Study

Graph Neural Networks (GNNs) have become essential tools for learning on relational data, yet the performance of a single GNN is often limited by the heterogeneity present in real-world graphs. Recent advances in Mixture-of-Experts (MoE) frameworks demonstrate that assembling multiple, explicitly diverse GNNs with distinct generalization patterns can significantly improve performance. In this work, we present the first systematic empirical study of expert-level diversification techniques for GNN ensembles. Evaluating 20 diversification strategies -- including random re-initialization, hyperparameter tuning, architectural variation, directionality modeling, and training data partitioning -- across 14 node classification benchmarks, we construct and analyze over 200 ensemble variants. Our comprehensive evaluation examines each technique in terms of expert diversity, complementarity, and ensemble performance. We also uncovers mechanistic insights into training maximally diverse experts. These findings provide actionable guidance for expert training and the design of effective MoE frameworks on graph data. Our code is available at https://github.com/Hydrapse/bench-gnn-diversification.

cs.LG↗

arXiv2025

Accelerating Dynamic Image Graph Construction on FPGA for Vision GNNs

Vision Graph Neural Networks (Vision GNNs, or ViGs) represent images as unstructured graphs, achieving state of the art performance in computer vision tasks such as image classification, object detection, and instance segmentation. Dynamic Image Graph Construction (DIGC) builds image graphs by connecting patches (nodes) based on feature similarity, and is dynamically repeated in each ViG layer following GNN based patch (node) feature updates. However, DIGC constitutes over 50% of end to end ViG inference latency, rising to 95% at high image resolutions, making it the dominant computational bottleneck. While hardware acceleration holds promise, prior works primarily optimize graph construction algorithmically, often compromising DIGC flexibility, accuracy, or generality. To address these limitations, we propose a streaming, deeply pipelined FPGA accelerator for DIGC, featuring on chip buffers that process input features in small, uniform blocks. Our design minimizes external memory traffic via localized computation and performs efficient parallel sorting with local merge sort and global k way merging directly on streaming input blocks via heap insertion. This modular architecture scales seamlessly across image resolutions, ViG layer types, and model sizes and variants, and supports DIGC across diverse ViG based vision backbones. The design achieves high clock frequencies post place and route due to the statically configured parallelism minimizing critical path delay and delivers up to 16.6x and 6.8x speedups over optimized CPU and GPU DIGC baselines.

cs.DC↗

arXiv2025

Mixture of Thoughts: Learning to Aggregate What Experts Think, Not Just What They Say

Open-source Large Language Models (LLMs) increasingly specialize by domain (e.g., math, code, general reasoning), motivating systems that leverage complementary strengths across models. Prior multi-LLM approaches either (i) route a query to one or a few experts and generate independently, (ii) aggregate outputs from each model via costly multi-turn exchanges, or (iii) fuse weights into a single model-typically requiring architectural homogeneity. We introduce Mixture of Thoughts (MoT), a simple method for latent-level collaboration among heterogeneous experts under a global routing scheme. For each query, a lightweight router selects top-$K$ experts and designates a primary expert; uniformly placed interaction layers project hidden states into a shared latent space where the primary expert performs cross-attention over its active (selected) peers. Pre-trained experts remain frozen; only the router and the lightweight interaction layers are trained with a novel joint training objective that improves both the expert selection and inter-expert collaboration. Across five in-distribution (ID) and three out-of-distribution (OOD) benchmarks, MoT surpasses the current routing and aggregation-based state-of-the-art, Avengers, by $+0.38\%$ and $+2.92\%$, respectively. Further, MoT significantly outperforms the best-performing single model. It achieves this with single-pass inference, runtime comparable to routing baselines, and none of the overheads of iterative aggregation. MoT offers a simple latent-space mechanism for combining heterogeneous LLMs, a practical step toward broader multi-LLM collaboration. Our code is publicly available at https://github.com/jacobfa/mot.

cs.LG↗

arXiv2025

AMPED: Accelerating MTTKRP for Billion-Scale Sparse Tensor Decomposition on Multiple GPUs

Matricized Tensor Times Khatri-Rao Product (MTTKRP) is the computational bottleneck in sparse tensor decomposition. As real-world sparse tensors grow to billions of nonzeros, they increasingly demand higher memory capacity and compute throughput from hardware accelerators. In this work, we present AMPED, a multi-GPU parallel algorithm designed to accelerate MTTKRP on billion-scale sparse tensors. AMPED scales beyond the limits of a single GPU, meeting both the memory and performance requirements of large-scale workloads. We introduce a partitioning strategy combined with a dynamic load balancing scheme to distribute computation and minimize GPU idle time. On real-world billion-scale tensors, AMPED achieves a 5.1x geometric mean speedup in total execution time over state-of-the-art GPU baselines using 4 GPUs on a single CPU node.

cs.DC↗

arXiv2025

ScalableHD: Scalable and High-Throughput Hyperdimensional Computing Inference on Multi-Core CPUs

Hyperdimensional Computing (HDC) is a brain-inspired computing paradigm that represents and manipulates information using high-dimensional vectors, called hypervectors (HV). Traditional HDC methods, while robust to noise and inherently parallel, rely on single-pass, non-parametric training and often suffer from low accuracy. To address this, recent approaches adopt iterative training of base and class HVs, typically accelerated on GPUs. Inference, however, remains lightweight and well-suited for real-time execution. Yet, efficient HDC inference has been studied almost exclusively on specialized hardware such as FPGAs and GPUs, with limited attention to general-purpose multi-core CPUs. To address this gap, we propose ScalableHD for scalable and high-throughput HDC inference on multi-core CPUs. ScalableHD employs a two-stage pipelined execution model, where each stage is parallelized across cores and processes chunks of base and class HVs. Intermediate results are streamed between stages using a producer-consumer mechanism, enabling on-the-fly consumption and improving cache locality. To maximize performance, ScalableHD integrates memory tiling and NUMA-aware worker-to-core binding. Further, it features two execution variants tailored for small and large batch sizes, each designed to exploit compute parallelism based on workload characteristics while mitigating the memory-bound compute pattern that limits HDC inference performance on modern multi-core CPUs. ScalableHD achieves up to 10x speedup in throughput (samples per second) over state-of-the-art baselines such as TorchHD, across a diverse set of tasks ranging from human activity recognition to image classification, while preserving task accuracy. Furthermore, ScalableHD exhibits robust scalability: increasing the number of cores yields near-proportional throughput improvements.

cs.DC↗

arXiv2025

RECIPE-TKG: From Sparse History to Structured Reasoning for LLM-based Temporal Knowledge Graph Completion

Temporal Knowledge Graphs (TKGs) represent dynamic facts as timestamped relations between entities. TKG completion involves forecasting missing or future links, requiring models to reason over time-evolving structure. While LLMs show promise for this task, existing approaches often overemphasize supervised fine-tuning and struggle particularly when historical evidence is limited or missing. We introduce RECIPE-TKG, a lightweight and data-efficient framework designed to improve accuracy and generalization in settings with sparse historical context. It combines (1) rule-based multi-hop retrieval for structurally diverse history, (2) contrastive fine-tuning of lightweight adapters to encode relational semantics, and (3) test-time semantic filtering to iteratively refine generations based on embedding similarity. Experiments on four TKG benchmarks show that RECIPE-TKG outperforms previous LLM-based approaches, achieving up to 30.6\% relative improvement in Hits@10. Moreover, our proposed framework produces more semantically coherent predictions, even for the samples with limited historical context.

cs.LG↗

arXiv2025

SPECTRE: An FFT-Based Efficient Drop-In Replacement to Self-Attention for Long Contexts

Long-context transformers face significant efficiency challenges due to the quadratic cost of self-attention. However, many modern applications-from multi-turn dialogue to high-resolution vision-require contexts spanning tens of thousands of tokens. We introduce SPECTRE, a method that replaces each attention head with a fast real FFT, a content-adaptive spectral gate, and an inverse FFT, reducing per-layer complexity from $\mathcal{O}(L^{2})$ to $O(L\log L)$ while preserving the surrounding architecture. We extend this efficiency to autoregressive generation through our Prefix-FFT cache and enhance local feature representation with an optional wavelet module that adds negligible computational overhead. Our experiments demonstrate that SPECTRE operates up to 7$\times$ faster than FlashAttention-2 on 128k-token contexts while matching or exceeding baseline performance on PG-19 language modeling and ImageNet-1k classification tasks. SPECTRE achieves these improvements by adding fewer than 6\% parameters to the base model, making hundred-kilotoken context processing feasible on commodity GPUs without specialized hardware.

cs.LG↗

arXiv2025

LocAgent: Graph-Guided LLM Agents for Code Localization

Code localization--identifying precisely where in a codebase changes need to be made--is a fundamental yet challenging task in software maintenance. Existing approaches struggle to efficiently navigate complex codebases when identifying relevant code sections. The challenge lies in bridging natural language problem descriptions with the appropriate code elements, often requiring reasoning across hierarchical structures and multiple dependencies. We introduce LocAgent, a framework that addresses code localization through graph-based representation. By parsing codebases into directed heterogeneous graphs, LocAgent creates a lightweight representation that captures code structures (files, classes, functions) and their dependencies (imports, invocations, inheritance), enabling LLM agents to effectively search and locate relevant entities through powerful multi-hop reasoning. Experimental results on real-world benchmarks demonstrate that our approach significantly enhances accuracy in code localization. Notably, our method with the fine-tuned Qwen-2.5-Coder-Instruct-32B model achieves comparable results to SOTA proprietary models at greatly reduced cost (approximately 86% reduction), reaching up to 92.7% accuracy on file-level localization while improving downstream GitHub issue resolution success rates by 12% for multiple attempts (Pass@10). Our code is available at https://github.com/gersteinlab/LocAgent.

cs.SE↗