I. Case Overview

Faced with the common industry challenge where algorithm teams excel at academic papers yet struggle with patent drafting, coupled with the high technical bar for drafting AI algorithm claims, the ZYIP team deferred immediate drafting. Instead, we ran a week-long “translation workshop” alongside the client, repeatedly discussing the gaps between algorithm logic and patent frameworks. We broke down academic algorithm descriptions into patent-eligible implementation steps. Centered on the core innovation of unsupervised anomaly detection based on reconstruction error, we built end-to-end protection covering method workflows, model training, error calculation and electronic devices. The invention patent was successfully published. For the first time, the client’s algorithm engineers found the patent specification read like their own algorithm notes.

II. Case Details

1. Client Profile & Industry

A provider of on-board video monitoring and intelligent analysis solutions, specializing in commercial vehicle intelligent driving and Internet of Vehicles (IoV).

2. Client Pain Points & Challenges

(1) Algorithm engineers’ frustration with patents: The client’s algorithm team produced high-quality academic papers, but prior patent drafts failed to reflect their actual work, gradually eroding confidence in patent filings. (2) High drafting threshold for AI algorithms: The solution involves sophisticated technologies including multimodal fusion, unsupervised learning and probabilistic sampling. Ordinary attorneys struggle to follow the underlying algorithm logic. (3) Difficulties protecting unsupervised detection: The core innovation is reconstruction-error-based unsupervised anomaly detection. This “zero-sample” paradigm inherently makes it hard to define a clear scope of protection. (4) Dual academic and commercial expectations: The algorithm team required patents that reflect academic merit while safeguarding commercial interests, laying solid legal foundations for their innovation.

3. Solutions & Implementation

(1) A week-long “translation workshop” Rather than jumping into drafting, ZYIP worked intensively with the algorithm team for one week. We iteratively clarified the differences between algorithm mechanisms and patent claim structures, converting paper-style algorithm narratives into patentable procedural steps. This process demanded more effort than the drafting itself.

(2) Pinpoint the core inventive mechanism After fully grasping the technology, our team identified the core innovation: unsupervised anomaly detection based on reconstruction error. This directly addressed the difficulty of defining protection boundaries for the zero-sample approach.

(3) Full-chain layout with granular protection for key nodes We covered the complete chain: method steps, model training, error calculation and electronic devices. Dependent claims were crafted for critical algorithm nodes such as cross-attention processing and latent distribution parameter sampling. The scope protects the core algorithm while extending to specific implementation details.

(4) Deliver readable documents for engineers When the first draft was shared with the algorithm team, the engineers had no complaints about incomprehensible wording. For them, recognizing the document as a faithful representation of their work was the highest praise.

4. Deliverables & Outcomes

① Patent outcome: The invention patent for anomaly event detection was successfully published, implementing anomaly detection via multimodal fusion of video, audio and conditional vectors. ② Protection robustness: Anchored on unsupervised anomaly detection based on reconstruction error, the portfolio spans method workflows, model training, error calculation and electronic devices. Dependent claims cover key links including cross-attention processing and latent distribution parameter sampling, enclosing both core algorithms and implementation details. ③ High client recognition: Upon delivery, the algorithm engineer commented: “This is the first time I’ve read a patent document that feels like our own algorithm notes.” For a team once discouraged by poor-quality patents, this recognition speaks volumes. ④ Commercial and industrial value: The client presented this patent portfolio as core competitive evidence in multiple commercial vehicle intelligent driving tenders. As a representative software case in the emerging IoV sector, it provides a replicable IP layout roadmap for other AI algorithm enterprises.

III. Case Significance

This case demonstrates ZYIP’s technical depth and client-centric service philosophy in IP support for AI algorithms. The main difficulty of AI algorithm patent drafting lies not in the technology itself, but in translation. An attorney willing to unpack academic papers into valid claims alongside algorithm engineers outperforms countless template-based drafters. It proves ZYIP’s capability to convert research-heavy algorithm papers into enforceable rights that engineers understand and competitors cannot easily circumvent, turning the client’s core algorithms into tangible competitive strengths in the intelligent driving track.