arXiv subjects

Ponnuswamy Sadayappan

Publications and source records attributed to Ponnuswamy Sadayappan.

7 recordsLinked to original sources

arXiv2026

Lesion Overlap Ranks PET Segmentation Models but Does Not Certify Patient-Level Tumor Burden

Metabolic tumor volume (MTV) summarizes whole-body PET disease burden for prognostic stratification and treatment assessment. Automated MTV is useful only if its estimate is reliable for the individual patient, yet segmentation models are usually evaluated with overlap metrics such as Dice rather than burden error. We examined whether overlap metrics identify unreliable MTV estimates using 1,014 autoPET FDG PET/CT studies. We evaluated LesionTracer predictions held out from fine-tuning, compared five selected released models on 489 tumor-positive patients, and examined the findings in 597 PSMA studies. Across the five FDG models, ranking by higher mean Dice gave the same order as ranking by lower mean relative MTV error ($ρ=-1.00$, exact $p=0.017$). At the patient level, however, out-of-fold Dice tracked relative MTV error strongly ($ρ=-0.75$), yet 82 of 489 patients (16.8\%) fell on opposite sides of the Dice 0.80 and MTV-error 25\% thresholds. Within the training-exposed released models, correlations ranged from $-0.47$ to $-0.74$ over narrower, high-Dice ranges. Out of fold, opposing false-positive and false-negative volumes cancelled about one third of spatial error in net MTV. In 236 of 489 patients, each of the five models missed at least one lesion, though not necessarily the same lesion. Thus, agreement in average model rankings did not establish the reliability of individual MTV estimates or reveal shared lesion misses. PET segmentation benchmarks should report patient-level MTV error and missed-lesion counts alongside Dice and other overlap metrics.

cs.CV↗

arXiv2026

FLASHSWIN: Unlocking Large Windows and Dense Tokens in Swin Vision Transformers with Memory Efficient Attention

High-resolution vision backbones have long been forced to trade away local token density to afford larger receptive fields. Hierarchical Swin transformers impose this compromise because standard windowed attention materializes an $M^2\times M^2$ score matrix per window, incurring $O(M^4)$ memory as windows or token grids grow. Furthermore, Swin adds a learned relative-position bias elementwise to attention scores, requiring full materialization of the score matrix and its gradient. This keeps Swin and SwinV2 trapped in a small-window($M=8,16$), coarse-token regime with patch size $4\times4$ ($p=4$), limiting performance for fine-grained tasks. We introduce FLASHSWIN, which replaces standard windowed attention with a FlashAttention implementation that computes exact softmax attention without materializing the score matrix, reducing per-window memory from $O(M^4)$ to $O(M^2)$. This enables higher token density and larger receptive fields without inflating memory overhead. Training memory is flat across window sizes: at a $32\times32$ window, FLASHSWIN-T requires only $12.4$\,GB, unchanged from $8\times8$, compared to $70/90$\,GB for SwinV2/V1-T. However, applying FlashAttention directly to Swin creates a trade-off: bypassing the score matrix precludes Swin's additive relative-position bias, forfeiting spatial information in exchange for memory efficiency. FLASHSWIN restores position information as window-local learnable 2D RoPE, making large windows and dense token grids both affordable and accurate. At matched scale, FLASHSWIN-T outperforms Swin variants. With dense tokens and wide windows ($p=2,M=32$), the same Tiny model reaches $84.1\%$ ImageNet-1K, $44.1$ COCO box AP, and $47.28$ ADE20K mIoU---gains of $+1.3$, $+5.1$, and $+1.82$ over SwinV2-T at $M=16$, respectively. At fixed $M=32$, halving the patch size yields roughly $3\times$ larger gains in boundary quality than in mIoU.

cs.CV↗

arXiv2025

Fast Algorithms for Scheduling Many-body Correlation Functions on Accelerators

Computation of correlation functions is a key operation in Lattice quantum chromodynamics (LQCD) simulations to extract nuclear physics observables. These functions involve many binary batch tensor contractions, each tensor possibly occupying hundreds of MBs of memory. Performing these contractions on GPU accelerators poses the challenge of scheduling them as to optimize tensor reuse and reduce data traffic. In this work we propose two fast novel scheduling algorithms that reorder contractions to increase temporal locality via input/intermediate tensor reuse. Our schedulers take advantage of application-specific features, such as contractions being binary and locality within contraction trees, to optimize the objective of minimizing peak memory. We integrate them into the LQCD analysis software suite Redstar and improve time-to-solution. Our schedulers attain upto 2.1x improvement in peak memory, which is reflected by a reduction of upto 4.2x in evictions, upto 1.8x in data traffic, resulting in upto 1.9x faster correlation function computation time.

cs.DC↗

arXiv2025

Optimizing Data Distribution and Kernel Performance for Efficient Training of Chemistry Foundation Models: A Case Study with MACE

Chemistry Foundation Models (CFMs) that leverage Graph Neural Networks (GNNs) operating on 3D molecular graph structures are becoming indispensable tools for computational chemists and materials scientists. These models facilitate the understanding of matter and the discovery of new molecules and materials. In contrast to GNNs operating on a large homogeneous graphs, GNNs used by CFMs process a large number of geometric graphs of varying sizes, requiring different optimization strategies than those developed for large homogeneous GNNs. This paper presents optimizations for two critical phases of CFM training: data distribution and model training, targeting MACE - a state-of-the-art CFM. We address the challenge of load balancing in data distribution by formulating it as a multi-objective bin packing problem. We propose an iterative algorithm that provides a highly effective, fast, and practical solution, ensuring efficient data distribution. For the training phase, we identify symmetric tensor contraction as the key computational kernel in MACE and optimize this kernel to improve the overall performance. Our combined approach of balanced data distribution and kernel optimization significantly enhances the training process of MACE. Experimental results demonstrate a substantial speedup, reducing per-epoch execution time for training from 12 to 2 minutes on 740 GPUs with a 2.6M sample dataset.

cs.DC↗

arXiv2025

TopV: Compatible Token Pruning with Inference Time Optimization for Fast and Low-Memory Multimodal Vision Language Model

Vision-Language Models (VLMs) demand substantial computational resources during inference, largely due to the extensive visual input tokens for representing visual information. Previous studies have noted that visual tokens tend to receive less attention than text tokens, suggesting their lower importance during inference and potential for pruning. However, their methods encounter several challenges: reliance on greedy heuristic criteria for token importance and incompatibility with FlashAttention and KV cache. To address these issues, we introduce \textbf{TopV}, a compatible \textbf{TO}ken \textbf{P}runing with inference Time Optimization for fast and low-memory \textbf{V}LM, achieving efficient pruning without additional training or fine-tuning. Instead of relying on attention scores, we formulate token pruning as an optimization problem, accurately identifying important visual tokens while remaining compatible with FlashAttention. Additionally, since we only perform this pruning once during the prefilling stage, it effectively reduces KV cache size. Our optimization framework incorporates a visual-aware cost function considering factors such as Feature Similarity, Relative Spatial Distance, and Absolute Central Distance, to measure the importance of each source visual token, enabling effective pruning of low-importance tokens. Extensive experiments demonstrate that our method outperforms previous token pruning methods, validating the effectiveness and efficiency of our approach.

cs.CV↗

arXiv2023

Automating GPU Scalability for Complex Scientific Models: Phonon Boltzman Transport Equation

Heterogeneous computing environments combining CPU and GPU resources provide a great boost to large-scale scientific computing applications. Code generation utilities that partition the work into CPU and GPU tasks while considering data movement costs allow researchers to more quickly and easily develop high-performance solutions, and make these resources accessible to a larger user base. We present developments for a domain-specific language (DSL) and code generation framework for solving partial differential equations (PDEs). These enhancements facilitate GPU-accelerated solution of the Boltzmann transport equation (BTE) for phonons, which is the governing equation for simulating thermal transport in semiconductor materials at sub-micron scales. The solution of the BTE involves thousands of coupled PDEs as well as complicated boundary conditions and nonlinear processing at each time step. These developments enable the DSL to generate configurable hybrid GPU/CPU code that couples accelerated kernels with user-defined code. We observed performance improvements of around 18X compared to a CPU-only version produced by this same DSL with minimal additional programming effort.

cs.CE↗

arXiv2011

Fast Sparse Matrix-Vector Multiplication on GPUs: Implications for Graph Mining

Scaling up the sparse matrix-vector multiplication kernel on modern Graphics Processing Units (GPU) has been at the heart of numerous studies in both academia and industry. In this article we present a novel non-parametric, self-tunable, approach to data representation for computing this kernel, particularly targeting sparse matrices representing power-law graphs. Using real data, we show how our representation scheme, coupled with a novel tiling algorithm, can yield significant benefits over the current state of the art GPU efforts on a number of core data mining algorithms such as PageRank, HITS and Random Walk with Restart.

math.NA↗