arXiv subjects

Gaurav Agarwal

Publications and source records attributed to Gaurav Agarwal.

6 recordsLinked to original sources

arXiv2026

How Much of a Real Workload Can LLM-Generated GPU Kernels Actually Reach?

Language models can now write GPU kernels that outperform PyTorch. We evaluate five model configurations on KernelBench level 1 and find that a frontier model produces correct kernels for 91.1% of problems and independently verified speedups on 22 of 56, including three convolutions, with a median of 1.235x. Open-weights models are far behind: the best reaches 30.4% correct with three verified speedups and solves zero convolutions. We then ask a question the literature does not: what fraction of a real model's wall clock do such kernels govern? Profiling seven workloads across three domains, we find the addressable fraction ranges from 8.9% to 58.2%. On transformers, 80-86% of runtime is spent in cuBLAS GEMM and FlashAttention, bounding realistic end-to-end improvement at roughly 1%, and the fraction shrinks with model scale. On recommenders it is 58.2%, concentrated in a single embedding kernel. We introduce DLRM-Bench, 12 recommender kernel problems in KernelBench format, and measure a 41.7% win rate at a 1.552x median there, projecting 8.63% end-to-end. Separately, we show that KernelBench's correctness check (torch.allclose with an absolute tolerance) is satisfied by a tensor of zeros on 4 of 60 level-1 problems. Two kernels in our own results exploited this before we detected them, including one scored at 283x that wrote 0.3% of its output buffer. We propose scale-invariant replacements and release all 879 evaluations.

cs.DC↗

arXiv2026

Fidelity-Aware Frequency Allocation and Transpilation Co-Design for Tunable Coupler Quantum Systems

Frequency crowding is a fundamental limitation in superconducting quantum architectures, particularly in tunable-coupler systems. We present a framework that explicitly models both coherent spectator-induced errors and incoherent lifetime effects through an error budgeting approach. Using this model, we analyze how frequency crowding impacts gate fidelity as module size and connectivity scale, and formulate a constrained optimization problem to assign qubit and coupler frequencies under realistic separation and hardware constraints. We demonstrate scalable frequency allocation strategies that minimize spectator-induced errors. We further show that increasing qubit count and coupling density within a module leads to a fidelity-connectivity tradeoff. To explore the benefits at the system scale, we have developed a noise-aware transpilation approach called FINESSE, which minimizes error by selecting high-fidelity paths that satisfy connectivity via SWAP insertion while jointly optimizing downstream gate execution. We demonstrate this physics-informed architecture-transpilation co-design approach for a SNAIL-based third-order coupler that natively realizes the $\sqrt{iSWAP}$ basis with frequency aware gate fidelities. On SNAIL architectures, FINESSE achieves an average 8.9% reduction in log-infidelity cost and 6.8% reduction in circuit depth vs. SABRE. We also compare results on IBM Brisbane's architecture.

quant-ph↗

arXiv2025

Spectator-Aware Frequency Allocation in Tunable-Coupler Quantum Architectures

This paper addresses frequency crowding in SNAIL-based superconducting quantum modules. First, we present design constraints by describing a physical model for realizable gates within a module, and building a fidelity model using error budgeting derived from device characteristics. Second, we tackle the allocation problem by analyzing the impact of frequency crowding on gate fidelity as the radix of the module increases. We explore whether the heuristic gate fidelity can be optimized with a discrete set of qubit frequencies while adhering to defined separation thresholds. By leveraging a combination of analytical and numerical techniques, we demonstrate scalable frequency allocation strategies that minimize spectator-induced errors. Our results further indicate that removing edges leads to improved gate fidelities while maintaining sufficient connectivity, suggesting that edge density is not a limiting factor for NISQ-scale benchmarks. The findings have implications for designing robust, high-fidelity quantum systems with practical constraints on hardware and connectivity.

quant-ph↗

arXiv2025

Fast Machine Learning for Quantum Control of Microwave Qudits on Edge Hardware

Quantum optimal control is a promising approach to improve the accuracy of quantum gates, but it relies on complex algorithms to determine the best control settings. CPU or GPU-based approaches often have delays that are too long to be applied in practice. It is paramount to have systems with extremely low delays to quickly and with high fidelity adjust quantum hardware settings, where fidelity is defined as overlap with a target quantum state. Here, we utilize machine learning (ML) models to determine control-pulse parameters for preparing Selective Number-dependent Arbitrary Phase (SNAP) gates in microwave cavity qudits, which are multi-level quantum systems that serve as elementary computation units for quantum computing. The methodology involves data generation using classical optimization techniques, ML model development, design space exploration, and quantization for hardware implementation. Our results demonstrate the efficacy of the proposed approach, with optimized models achieving low gate trace infidelity near $10^{-3}$ and efficient utilization of programmable logic resources.

quant-ph↗

arXiv2019

Deep Reinforcement Learning for Single-Shot Diagnosis and Adaptation in Damaged Robots

Robotics has proved to be an indispensable tool in many industrial as well as social applications, such as warehouse automation, manufacturing, disaster robotics, etc. In most of these scenarios, damage to the agent while accomplishing mission-critical tasks can result in failure. To enable robotic adaptation in such situations, the agent needs to adopt policies which are robust to a diverse set of damages and must do so with minimum computational complexity. We thus propose a damage aware control architecture which diagnoses the damage prior to gait selection while also incorporating domain randomization in the damage space for learning a robust policy. To implement damage awareness, we have used a Long Short Term Memory based supervised learning network which diagnoses the damage and predicts the type of damage. The main novelty of this approach is that only a single policy is trained to adapt against a wide variety of damages and the diagnosis is done in a single trial at the time of damage.

cs.LG↗

arXiv2008

Modularity-Maximizing Network Communities via Mathematical Programming

In many networks, it is of great interest to identify "communities", unusually densely knit groups of individuals. Such communities often shed light on the function of the networks or underlying properties of the individuals. Recently, Newman suggested "modularity" as a natural measure of the quality of a network partitioning into communities. Since then, various algorithms have been proposed for (approximately) maximizing the modularity of the partitioning determined. In this paper, we introduce the technique of rounding mathematical programs to the problem of modularity maximization, presenting two novel algorithms. More specifically, the algorithms round solutions to linear and vector programs. Importantly, the linear programing algorithm comes with an a posteriori approximation guarantee: by comparing the solution quality to the fractional solution of the linear program, a bound on the available "room for improvement" can be obtained. The vector programming algorithm provides a similar bound for the best partition into two communities. We evaluate both algorithms using experiments on several standard test cases for network partitioning algorithms, and find that they perform comparably or better than past algorithms.

physics.data-an↗