Saanvi Kataria | Searcharxiv

arXiv2026

Three Months in the Life of Cloud Quantum Computing

Quantum Computing (QC) has evolved from a few custom quantum computers, which were only accessible to their creators, to an array of commercial quantum computers that can be accessed on the cloud by anyone. Accessing these cloud quantum computers requires a complex chain of tools that facilitate connecting, programming, simulating algorithms, estimating resources, submitting quantum computing jobs, retrieving results, and more. Some steps in the chain are hardware dependent and subject to change as both hardware and software tools, such as available gate sets and optimizing compilers, evolve. Understanding the trade-offs inherent in this process is essential for evaluating the power and utility of quantum computers. ARLIS has been systematically investigating these environments to understand these complexities. The work presented here is a detailed summary of three months of using such quantum programming environments. We show metadata obtained from these environments, including the connection metrics to the different services, the execution of algorithms, the testing of the effects of varying the number of qubits, comparisons to simulations, execution times, and cost. Our objective is to provide concrete data and insights for those who are exploring the potential of quantum computing. It is not our objective to present any new algorithms or optimize performance on any particular machine or cloud platform; rather, this work is focused on providing a consistent view of a single algorithm executed using out-of-the-box settings and tools across machines, cloud platforms, and time. We present insights only available from these carefully curated data.

quant-ph↗

arXiv2025

Domain-Robust Marine Plastic Detection Using Vision Models

Marine plastic pollution is a pressing environmental threat, making reliable automation for underwater debris detection essential. However, vision systems trained on one dataset often degrade on new imagery due to domain shift. This study benchmarks models for cross-domain robustness, training convolutional neural networks - CNNs (MobileNetV2, ResNet-18, EfficientNet-B0) and vision transformers (DeiT-Tiny, ViT-B16) on a labeled underwater dataset and then evaluates them on a balanced cross-domain test set built from plastic-positive images drawn from a different source and negatives from the training domain. Two zero-shot models were assessed, CLIP ViT-L14 and Google's Gemini 2.0 Flash, that leverage pretraining to classify images without fine-tuning. Results show the lightweight MobileNetV2 delivers the strongest cross-domain performance (F1 0.97), surpassing larger models. All fine-tuned models achieved high Precision (around 99%), but differ in Recall, indicating varying sensitivity to plastic instances. Zero-shot CLIP is comparatively sensitive (Recall around 80%) yet prone to false positives (Precision around 56%), whereas Gemini exhibits the inverse profile (Precision around 99%, Recall around 81%). Error analysis highlights recurring confusions with coral textures, suspended particulates, and specular glare. Overall, compact CNNs with supervised training can generalize effectively for cross-domain underwater detection, while large pretrained vision-language models provide complementary strengths.

cs.CV↗