Abstracts of published research. Full papers are at the publisher.
← Back to Documentation HubProposes an intent-aware control plane whose central contribution is the decision artifact: a durable, versioned, auditable record of how an intent became an executable plan — which policies applied, which capabilities were evaluated, which alternatives were rejected, and why. A prototype of the decision layer agrees with the OPA Gatekeeper constraint library's published verdicts on 42 of 42 encodable cases; the context-signal and ranking stages remain design claims.
arXiv:2609.21299 [cs.DC], v1, 18 September 2026. DOI 10.48550/arXiv.2609.21299. CC BY 4.0.
Reports a small model trained under Brain API to judge which capability of an Agentic Dataset a request is for, among the capabilities policy has already admitted; policy decides what may run, and every decision is recorded with the model's version. On generated dataset families never seen in training, the trained 1.5B model selects the right capability 98.1% of the time against 92.6% for the same base model prompted, and reaches 98.1% against 98.2% for a 7B reference at a fifth of its median latency. All 3,393 decision records re-derive from their recorded inputs to the same bytes.
Results note and carousel, Zenodo, 23 October 2026. DOI 10.5281/zenodo.23586630. CC BY 4.0.
Supplies the half of Croissant's data use conditions that no version specifies — how they are evaluated: an additive profile lets a dataset declare the operations it admits and the conditions under which it admits them, over five operators with a complete decision procedure, so a gate decides from the descriptor alone and records what it checked. On descriptors that gated a real nf-core pipeline, profile decisions match the native record for record at 11.7 µs added cost against a 119 µs decision.
arXiv:2609.19640 [cs.LG], v1, 17 September 2026. DOI 10.48550/arXiv.2609.19640.
Slides for a Birds-of-a-Feather session on what it would take to put a quantum step inside a regulated biopharmaceutical pipeline: what evidence must accompany a quantum result before a regulated workflow can evaluate it, which biological use cases are credible on noisy intermediate-scale devices and which depend on fault tolerance, and which governance requirements hold in both regimes.
BOF-2099, 14 September 2026. Slides on Zenodo, DOI 10.5281/zenodo.22841352. CC BY 4.0.
Introduces a policy-aware supervisory architecture for heterogeneous biomedical manipulators in which task intents, lifecycle states, policy decisions, refusal rationale and telemetry traces are first-class runtime artifacts, so an unsafe intent is refused before any planner or driver runs; a simulated plate transfer across a Cartesian pipettor and a 6-DOF arm shows the same intent-level workflow on both, with complete trace reconstruction.
In the conference proceedings; IEEE Xplore record pending.
Introduces dataset descriptors as a control-plane abstraction for autonomous biomedical pipelines: structured, versioned specifications that bind a dataset to its policies, permissible actions, and observability requirements, enabling decentralized policy enforcement and real-time auditability in regulated research environments.
In the conference proceedings; IEEE Xplore record pending.
Presents a control-plane architecture for Big Data as a Service that models workflows as adaptive, policy-aware service entities rather than static DAGs, combining a dual-record execution model adapted from pharmaceutical batch manufacturing with workflow-level semantic caching; on clustered workloads the cache reduces backend requests by 81% and median latency by 93%.
IEEE, pp. 125–129. DOI 10.1109/BigDataService70481.2026.00025
Introduces Agentic Datasets as a standalone engineering control-plane abstraction in which datasets are elevated from passive storage artifacts to first-class stateful entities encoding intent, policy, and lifecycle semantics across heterogeneous data-intensive infrastructure.
IEEE, pp. 326–332. DOI 10.1109/CCECE68150.2026.11610344
DEVS-grounded control-plane formulation of agentic datasets, framing dataset evolution as discrete-event behavior and using discrete-event simulation to evaluate policy, observability, and lifecycle semantics for data-intensive systems.
Work-in-progress paper, 4 pp, SCS Digital Library.
Journal article on tunnelling photoionization in a two-color (bichromatic) laser field.
Vol. 34, no. 3, pp. 363–368, 2001. DOI 10.1088/0953-4075/34/3/313