CVD

Classical Value of Information (VoI) methods typically assume that information value is additive or ignore the mutual influence among consultation points, whereas CVD explicitly models the sequential dependence and interaction effects of consultation decisions through joint value learning and continuation-aware value learning.

CVD was experimented on in five batch settings and three sequential settings, using two frontier executors.

Experimental results show that independent consultation value (i.e., value estimates that do not account for the mutual influence among consultation points) remains competitive across multiple settings.

Compared with direct prompting and the Value of Information method suited to specific scenarios, CVD performs better across a broad range of scenarios.

CVD remains robust when rollout supervision is substantially reduced, and can achieve near-lossless approximate allocation at low search cost.

CVD's consultation value predictor needs to be retrained, or at least recalibrated, after the executor is changed.

The paper proposes a solution named 'Consultation Value Distillation' (CVD), which uses Monte Carlo executor rollouts to estimate consultation value offline and then distills it into a lightweight predictor for budget allocation at deployment time.

The core innovation of CVD lies in joint value learning and continuation-aware value learning: joint value learning accounts for the mutual influence among multiple consultation points, and continuation-aware value learning accounts for the sequential nature of consultation decisions.

The common time window for coordinated vulnerability disclosure (CVD) is 90 days, i.e., the disclosure policy promoted by Google Project Zero.

When multiple vendors adopt a similar 90-day window and they all happen to expire in the same period, a short-term spike in vulnerability announcements appears in public channels, which is easily misread as a sudden increase in vulnerabilities.

The coordinated vulnerability disclosure (CVD) mechanism aims to shorten the patch gap and reduce information asymmetry, but early leaks can still occur in practice.

In open-source settings, CVD is especially complex, because a single library may be indirectly depended on by thousands of projects, maintainers need to notify downstream parties layer by layer, and the disclosure policies and response capabilities of each project vary widely.

In the coordinated vulnerability disclosure (CVD) process, researchers first report the vulnerability to the vendor or CERT/CC, granting a 90-to-120-day remediation window, and disclose details only after the period expires.

The mass-production bottlenecks for 2D semiconductors are concentrated at three levels: thin-film growth processes (with CVD as the mainstream approach), thermal budget compatibility with back-end-of-line (BEOL) interconnect processes, and contact resistance issues.

Future research directions include further reducing rollout cost, adapting to dynamic environments, and handling multi-human consultation scenarios.