Intuit (United States)

Tensor network structure search (TN-SS) aims to automatically discover optimal network topologies and rank configurations for efficient tensor decomposition in high-dimensional data representation. Despite recent advances, existing TN-SS methods face significant limitations in computational tractability, structure adaptivity, and optimization robustness across diverse tensor characteristics. They struggle with three key challenges: single-scale optimization missing multi-scale structures, discrete search spaces hindering smooth structure evolution, and separated structure-parameter optimization causing computational inefficiency. We propose RGTN (Renormalization Group guided Tensor Network search), a physics-inspired framework transforming TN-SS via multi-scale renormalization group flows. Unlike fixed-scale discrete search methods, RGTN uses dynamic scale-transformation for continuous structure evolution across resolutions. Its core innovation includes learnable edge gates for optimization-stage topology modification and intelligent proposals based on physical quantities like node tension measuring local stress, and edge information flow quantifying connectivity importance. Starting from low-complexity coarse scales and refining to finer ones, RGTN finds compact structures while escaping local minima via scale-induced perturbations. Extensive experiments on light field data, high-order synthetic tensors, and video completion tasks show RGTN achieves state-of-the-art compression ratios and runs 4-600 times faster than existing methods, validating the effectiveness of our physics-inspired approach.

Endoscopic Mini or Less Open Sublay (eMILOS) is a transhernial endoscopically assisted approach for anterior abdominal wall hernia repair that has demonstrated excellent long-term outcomes. Performed through a small skin incision directly over the hernia, the eMILOS technique minimizes additional access trauma, but remains technically demanding with conventional instruments. This study investigates the preclinical feasibility of adapting the da Vinci SP system to perform the transhernial SP MILOS technique. This preclinical study evaluated the technical feasibility of performing daVinci SP MILOS hernia repair using human cadavers. The da Vinci SP robotic platform and its standard instruments were employed. Procedural steps, technical considerations, limitations, and potential advantages were systematically documented. The procedure was completed through a 1.8-cm skin incision. Correct placement of the Access Port (AP) and defining the custom remote center (CRC) at the posterior fascial edge was essential for optimal trocar orientation and to limit collisions with the incision. The AP can be positioned below the skin or below the fascia depending on work flow and surgeon preference. The retrorectus space was developed robotically from xiphoid to pubis, with full 360° transhernial exposure. Transversus abdominis release and lateral dissection into the preperitoneal and pretransversalis planes were successfully achieved. In this preclinical cadaveric study, the SP MILOS technique provided 360° transhernial access to the retrorectus space without platform repositioning. Our initial experience supports its feasibility as a new single-port robotic option for ventral hernia repair, eliminating access trauma to the abdominal wall through a single, minimal skin incision.

Lightweight cryogenic propellant storage solutions with zero or near zero boil-off represent a critical need for NASA in both aeronautics and space technology missions. Applications for composite cryogen storage tanks range from hydrogen powered aircraft to long-duration in-space and on-surface missions. Over the past several decades, NASA and industry partners have supported several notable composite cryotank development programs documenting lessons learned with respect to materials and manufacturing. This paper specifically reports the results of material screening efforts funded by both NASA’s Aeronautics Research Mission Directorate and Space Technology Mission Directorate to evaluate thermoset and thermoplastic composites for cryogen storage. Although the end applications differ, the fundamental material requirements across the two projects are the same. This paper details characterization of candidate thermoplastic and thermoset composite materials for cryogen storage, with test data generated (1) at -196oC and (2) following 100 thermal cycles between room temperature (21oC) and LN2 temperature (-196oC).

Abstract Deploying large language models (LLMs) in clinical settings is limited by security, reliability, latency, and accessibility concerns that favor smaller, on-device or on-premise models. However, these smaller models may struggle to meet accuracy requirements. While fine-tuning and retrieval-augmented generation (RAG) can improve domain-specific accuracy, these methods require additional labeled data, technical skill, and infrastructure. In contrast, test-time scaling —allocating extra token-budget during inference—offers a training-free alternative to increasing accuracy. However, the trade-offs between these strategies and their interaction with model size remain poorly understood for medical reasoning. To address this gap, we compare three approaches—test-time scaling, fine-tuning, and context grounding—using the Gemma and MedGemma family of LLMs (Gemma-3 1B, Gemma-3 4B, Gemma-3 27B, MedGemma-4B, and MedGemma 27B) and evaluate these systems across common biomedical question-answering (QA) datasets and a set of recently released medical exam questions with the performance of practicing clinicians available for comparison. We test baseline prompts (direct answer, Chain-of-Thought, and self-consistency) while introducing a new prompting method we call “prompt-chaining for continuous reflection” (PCCR) that forces inference time minimum token-generation budgets. We assess accuracy and tokens-generated, allowing us to investigate the accuracy–efficiency trade-offs across prompting, context-grounding, fine-tuning, and model scales. We discover equivalency points where smaller models perform comparably to larger ones with increased reasoning budgets, context-grounding, or fine-tuning. We also find inflection points where context-grounding and test-time scaling used together lead to degrading performance. Using these empirical results, we formulate a general framework with equations to balance cost-benefit trade-offs when engineering LLM-based systems for medical reasoning and QA. We recommend generalizable configurations, designs, and patterns to achieve accuracy and efficiency objectives for example use-cases relevant to healthcare organizations. Author summary When doctors and hospitals want to use artificial intelligence for medical tasks, they face difficult choices. The most capable AI systems are expensive to run and require sending sensitive patient and hospital data to external servers. Smaller systems that can run locally are more practical but may be less accurate. Our research asked: with what configurations can we make smaller AI systems perform as well as larger ones for medical reasoning? We tested five AI models of different sizes, including both generalist and medically-specialized models, on thousands of medical questions using various prompting strategies, including a new and very simple method we developed that encourages the AI to reason more extensively. We discovered several surprising findings which we describe in detail. Most importantly, we found that smaller, cheaper models can match larger ones when given the right combination of prompting strategy, specialized training, and supporting information. We translated these findings into practical guidelines that can help choose AI configurations that balance accuracy, speed, and cost. Our framework could help make medical AI more accessible to institutions with limited computational resources while achieving the highest possible accuracy.

Surgical video review is essential for minimally invasive surgical training, but manual annotation of surgical steps is time-consuming and limits scalability. We propose a weakly supervised pre-training framework that leverages unannotated or heterogeneously labeled surgical videos to improve automated surgical step recognition. We evaluate three types of weak labels derived from unannotated datasets: (1) surgical phases from the same or other procedures, (2) surgical steps from different procedure types, and (3) intraoperative time progression. Using datasets from four robotic-assisted procedures (sleeve gastrectomy, hysterectomy, cholecystectomy, and radical prostatectomy), we simulate real-world annotation scarcity by varying the proportion of available step annotations ( 0.25, 0.5, 0.75, 1.0). We benchmark the performance of a 2D CNN model trained with and without weak label pre-training. Pre-training with surgical phase labels-particularly from the same procedure type (PHASE-WITHIN)-consistently improved step recognition performance, with gains up to 6.4 f1-score points over standard ImageNet-based models under limited annotation conditions ( = 0.25 on SLG). Cross-procedure step pre-training was beneficial for some procedures, and time-based labels provided moderate gains depending on procedure structure. Label efficiency analysis shows the baseline model would require labeling an additional 30-60 videos at = 0.25 to match the performance achieved by the best weak-pretraining strategy across procedures. Weakly supervised pre-training offers a practical strategy to improve surgical step recognition when annotated data is scarce. This approach can support scalable feedback and assessment in surgical training workflows where comprehensive annotations are infeasible.

Financial institutions are increasingly dealing with difficulties in preserving transaction integrity due to escalating digital fraud and extensive cloud-based operations. Conventional machine learning and rule-based detection systems frequently have latency in replies, restricted adaptability, and an absence of verifiable audit trails. The proposed study presents a Hybrid Multi-Agent Reinforcement Learning (MARL) and Blockchain framework to overcome existing limitations, integrating adaptive multi-agent coordination for real-time fraud detection with a permissioned blockchain ledger to guarantee secure, tamper-proof validation of agent decisions. MARL agents adaptively acquire optimal detection strategies through incentive shaping and inter-agent communication, while blockchain consensus and smart contracts substantiate decision proofs for regulatory adherence. The experimental evaluation of the Kaggle Financial Transactions dataset demonstrates improved accuracy relative to baseline models: Accuracy 0.989, Precision 0.981, Recall 0.987, F1-Score 0.984, AUC-ROC 0.992, with negligible latency (1.8 ms/transaction). The comparative research reveals a persistent superiority over Random Forest, XGBoost, CNN- LSTM, and Transformer-based classifiers, showcasing robustness, scalability, and real-time application. The suggested study enhances the area by merging reinforcement learning adaptability with blockchain security, establishing a realistic, high-throughput, and reliable framework for cloud- based financial systems, while presenting a model for future research in agentic AI-driven fraud prevention.

In addition to immediate course-level outcomes, this work also motivates the need for sustained integration of DevSecOps practices into curricula, with future plans to conduct longitudinal studies following graduates into professional practice to evaluate retention and industry impact.

The rapid increase of multi-cloud adoptions and federated data ecosystems has upended the way enterprises manage and protect personally identifiable information. Although such designs enable scalability, flexibility, and business cross-industry cooperation, they are also facing new challenges in the privacy and security area, because of diverse hardware characteristics among countries, data transfer across borders, and different definitions of privacy handling. The old school perimeter-based controls, which worked in static and siloed setups, are pretty useless when you're always distributed, and often AI-driven operations get into gear. Towards that end, this article presents the referred to unified AI-powered Privacy Engineering framework, which incorporates federated learning, differential privacy, zero trust philosophy, and automated governance into the design and operation of new generation “cloud native” systems. The proposed framework will highlight 4 architectural layers: (i) Federated Data Integration - which secure collaboration is enabled without centralizing raw PII, (ii) Privacy-Enhancing Technologies (PETs) such as homomorphic encryption, secure enclave and differential privacy to maintain the confidentiality under distributed processing; (iii) AI-Assisted Governance and Compliance - a intelligent policy orchestration automates regulatory alignment with real-time data lineage tracking; and (iv) Zero-Trust Adaptive Security - that it necessitates continual verification and anomaly detection on multi-cloud environments. With the infusion of AI on every layer, the framework evolves from its original reactive compliance enforcement to proactive context-aware privacy management. The empirical validation is executed through cases in healthcare and finance. In healthcare, federated oncology models showed that PII leakage was reduced by 38% with performance comparable to near-baseline, and compliance report time went down by 42%. That is, we negotiated a 25% latency reduction in detecting anomalies for a global bank’s fraud detection pipeline and a 40% improvement in cross-border audit readiness in the financial domain. Benchmark analysis also reveals that AI-enabled privacy engineering brings: 25% reduction in integration errors; 30-40% speedup to secure deployment; and 20% increase in throughput for federated workflows. The findings demonstrate both the technical feasibility and the strategic necessity of AI-driven privacy engineering at a time when regulatory mandates, adversarial AI risks, and pressures from cross-border data sharing have converged. The results suggest that building privacy into the architecture of systems is now a necessity to maintain resilience, compliance, and trust in multi-cloud and federated ecosystems

Japanese document figures present distinctive challenges. They encompass charts, tables, and conceptual/explanatory (''ponchi-e'') diagrams; mix tategaki (vertical writing) and yokogaki (horizontal writing); compress information into small labels and furigana; and rely heavily on layout to convey meaning. These properties make them both difficult and business-critical to understand. Motivated by this reality, SB Intuitions is developing Sarashina2-Vision, a Vision-Language Model (VLM) built on our Japanese LLM Sarashina2 and centered on handling such Japanese document figures. In this talk, we share lessons from development—data selection, instruction design, and evaluation—and outline future directions for more reliable Japanese figure understanding, situating our approach within the latest advances in figure-centric VLMs and incorporating insights from Japanese VLM research.