DCT
3:25-cv-08107
Google LLC v. Kmizra LLC
I. Executive Summary and Procedural Information
- Parties & Counsel:
- Plaintiff: Google LLC (Delaware)
- Defendant: K.Mizra LLC (Delaware)
- Plaintiff’s Counsel: Jones Day
- Case Identification: 3:25-cv-08107, N.D. Cal., 09/24/2025
- Venue Allegations: Plaintiff alleges venue is proper in the Northern District of California because Defendant is subject to personal jurisdiction there, has purposefully directed licensing and enforcement activities at companies in the district, including Google, and a substantial part of the events giving rise to the claim occurred in the district.
- Core Dispute: Plaintiff Google seeks a declaratory judgment that its Nest smart home products and Vertex AI machine learning platform do not infringe Defendant K.Mizra's patents related to wireless network initialization and hyperparameter estimation, respectively.
- Technical Context: The technologies at issue involve foundational methods for establishing ad-hoc wireless networks, critical for IoT devices, and for optimizing machine learning model performance, a key function in AI development platforms.
- Key Procedural History: The complaint alleges that this declaratory judgment action was precipitated by Defendant’s threats to sue Plaintiff for infringement of the patents-in-suit. This includes providing Plaintiff with a draft amended complaint in a separate, ongoing litigation in the Western District of Texas that would have added the current patents-in-suit. Defendant is characterized as a non-practicing entity whose business model is based on patent licensing and litigation.
Case Timeline
| Date | Event |
|---|---|
| 2005-12-30 | U.S. Patent No. 8,144,717 Priority Date |
| 2007-04-25 | U.S. Patent No. 8,438,120 Priority Date |
| 2012-03-27 | U.S. Patent No. 8,144,717 Issue Date |
| 2013-05-07 | U.S. Patent No. 8,438,120 Issue Date |
| 2022-06-06 | Defendant's CEO first directs communications to a Google employee |
| 2023-03-07 | Defendant's CEO communicates a request for Google to take a license |
| 2025-02-18 | Defendant files separate "Texas action" against Google on different patents |
| 2025-05-23 | Defendant's counsel requests leave to amend the Texas complaint to add the Patents-in-Suit |
| 2025-05-28 | Defendant provides a draft amended complaint alleging infringement by the Patents-in-Suit |
| 2025-06-09 | Defendant informs Plaintiff it has "decided against amending their complaint at this time" |
| 2025-06-16 | Defendant's CEO emails Plaintiff offering a license that includes the Patents-in-Suit |
| 2025-09-24 | Complaint for Declaratory Judgment filed |
II. Technology and Patent(s)-in-Suit Analysis
U.S. Patent No. 8,144,717 - "Initialization of a Wireless Communication Network"
The Invention Explained
- Problem Addressed: The patent addresses the challenge of efficiently and dynamically establishing communication routes in a wireless network of sensor stations without a pre-defined topology. Prior art methods were described as slow, complex, or resulting in inefficient routing with unnecessary branches. (’717 Patent, col. 2:1-3, 58-62).
- The Patented Solution: The invention discloses an orderly, hierarchical method for network formation. Devices ("stations") begin in a "not-associated state" and transmit "association request messages." Only the central "association unit" or another station that is already in an "associated state" can grant these requests. This creates a "wave of association that radiates from nodes to nodes," establishing a clear communication route for each new station back to the association unit by building upon the routes of already-associated stations. (’717 Patent, Abstract; col. 3:13-33).
- Technical Importance: This approach provided a simplified and structured method for self-configuring ad-hoc wireless networks, a key enabler for large-scale sensor deployments and early Internet of Things (IoT) applications where manual setup is impractical. (’717 Patent, col. 3:5-12).
Key Claims at a Glance
- The complaint identifies independent claim 11 as the asserted claim. (Compl. ¶36).
- Essential elements of claim 11 include:
- A station for a wireless network with a transmitter, receiver, and processor.
- The processor starts in a "not-associated state" and transmits association requests.
- The processor switches to an "associated state" upon receiving an association grant.
- The processor establishes an "associated route" to an association unit, where the route includes the source of the grant.
- If the grant source is not the association unit, the route is established using the "operating route between the source of the association grant and the association unit previously established."
- The processor is configured to transmit its own association grants to other stations, but only after it has switched to the associated state.
- The complaint notes that Defendant’s draft complaint treated claim 11 as representative of other claims. (Compl. ¶38).
U.S. Patent No. 8,438,120 - "Machine Learning Hyperparameter Estimation"
The Invention Explained
- Problem Addressed: The patent identifies the difficulty of determining optimal hyperparameters for machine learning classifiers. The search space for these parameters is often extremely large, making methods like simple grid searches inefficient. It notes that while the cross-entropy method is an efficient estimation technique, it is not readily applicable to hyperparameter tuning. (’120 Patent, col. 1:49-61).
- The Patented Solution: The invention proposes an iterative method to find optimal hyperparameters. In each iteration, the system (1) draws a random sample of potential hyperparameter vectors, (2) identifies the single hyperparameter vector that has produced the best performance result in the current or any previous iteration, and (3) updates the estimate of the target hyperparameter vector specifically using that single "best result" vector. This "elitist" approach ensures that the best-found solution is preserved and used to guide subsequent searches. (’120 Patent, Abstract; col. 2:13-26).
- Technical Importance: This method provides a more efficient and guided approach to hyperparameter optimization, a critical and often time-consuming step in developing high-accuracy machine learning models. (’120 Patent, col. 2:27-33).
Key Claims at a Glance
- The complaint identifies independent claim 1 as the asserted claim. (Compl. ¶44).
- Essential elements of claim 1 include:
- A method for determining hyperparameters of a classifier by iteratively producing an estimate of a target hyperparameter vector.
- Each iteration includes the steps of:
- drawing a random sample of hyperparameter vectors;
- updating the estimate of the target vector using the random sample; and
- selecting, from the random sample, a vector that produces a "best result in the present and any previous iterations."
- The claim specifies that the "step of updating the estimate... uses said hyperparameter vector producing the best result."
- The complaint notes that Defendant’s draft complaint treated claim 1 as representative of other claims. (Compl. ¶46).
III. The Accused Instrumentality
Product Identification
- For the ’717 Patent: Google's Nest line of products that support the Thread networking protocol, including the Nest Hub, Hub Max, Wifi Pro, and Google TV Streamer (collectively, the "Accused Thread Products"). (Compl. ¶38).
- For the ’120 Patent: Google's Vertex AI Studio, including its Vizier service (collectively, "Vertex AI"). (Compl. ¶46).
Functionality and Market Context
- The Accused Thread Products are smart home devices that use the Thread protocol to form a low-power, self-healing mesh network, allowing devices to communicate with each other and the internet. (Compl. ¶38).
- Vertex AI is a cloud-based platform for developing and deploying machine learning models. The Vizier service within Vertex AI is a black-box optimization service specifically used for tuning hyperparameters to improve model performance. (Compl. ¶46). The complaint contains a screenshot of Defendant's website listing its current litigations, including one against Google, which Plaintiff presents as evidence of Defendant's business model. (Compl. p. 3).
IV. Analysis of Infringement Allegations
U.S. Patent No. 8,144,717 Infringement Allegations
| Claim Element (from Independent Claim 11) | Alleged Infringing Functionality (per Google's non-infringement pleading) | Complaint Citation | Patent Citation |
|---|---|---|---|
| the processor circuit being configured to establish an associated route for communication with an association unit during subsequent communication... | The Accused Thread Products do not practice this limitation. | ¶39 | col. 15:5-13 |
| the associated route including a source of the association grant, and, if the source of the association grant is not the association unit, the operating route between the source of the association grant and the association unit previously established... | The Accused Thread Products do not establish routes using the specific hierarchical method of building upon a "previously established" route of the grant's source. | ¶39 | col. 15:5-13 |
Identified Points of Contention
- Technical Question: The central dispute may concern the specific routing mechanism of the Thread protocol as implemented in the Accused Thread Products. A key question is whether Thread establishes a new device's route by strictly adopting and extending the "previously established" route of the device that granted it network access, or whether it uses a more dynamic routing algorithm common in mesh networks that may establish a different, more optimal path. Google alleges the latter. (Compl. ¶39).
U.S. Patent No. 8,438,120 Infringement Allegations
| Claim Element (from Independent Claim 1) | Alleged Infringing Functionality (per Google's non-infringement pleading) | Complaint Citation | Patent Citation |
|---|---|---|---|
| updating the estimate of the target hyperparameter vector by using the random sample, and | Vertex AI does not practice this limitation. | ¶46 | col. 8:20-22 |
| [s]electing, from the random sample of hyperparameter vectors, a hyperparameter vector producing a best result in the present and any previous iterations, and wherein the step of updating the estimate of the target hyperparameter vector uses said hyperparameter vector producing the best result. | Vertex AI does not perform the updating step by using the single "best result" hyperparameter vector as required by the claim. | ¶46 | col. 8:23-28 |
Identified Points of Contention
- Scope Question: The analysis will likely focus on the meaning of "uses said hyperparameter vector producing the best result." Does this require the "best" vector to be the sole or primary input for the update calculation, or can the update "use" it in a broader sense, such as part of a weighted average of multiple high-performing vectors?
- Technical Question: What evidence will show how the Vertex AI / Vizier algorithm actually performs the "updating" step? The dispute will turn on whether the algorithm's update function is driven specifically and solely by the single best-performing vector found to date, or if it incorporates information from other vectors in the random sample in a manner distinct from the claimed method. Google alleges its products do not practice this limitation. (Compl. ¶46).
V. Key Claim Terms for Construction
U.S. Patent No. 8,144,717
- The Term: "the operating route between the source of the association grant and the association unit previously established"
- Context and Importance: This term is critical to defining the required method of route formation. Infringement appears to hinge on whether a newly associated station's route is constructed by appending itself to the specific, static, pre-existing path of the station that granted it access. Practitioners may focus on this term because modern mesh protocols often use dynamic routing tables rather than the rigid, hierarchical path-building described in the patent.
- Intrinsic Evidence for Interpretation:
- Evidence for a Narrower Interpretation: The specification describes a "wave of association that radiates from nodes to nodes starting from the association unit" and that routes are established "using routes that have already been defined." (’717 Patent, col. 3:16-19, 30-32). This language may support a narrow construction requiring a direct, daisy-chained route formation.
- Evidence for a Broader Interpretation: A party might argue that any route that uses the granting station as its next hop inherently "uses" that station's "previously established" connection to the network, even if the ultimate path to the association unit is recalculated or differs from the granting station's own path.
U.S. Patent No. 8,438,120
- The Term: "updating the estimate ... uses said hyperparameter vector producing the best result"
- Context and Importance: The definition of this term is central to the non-infringement argument. The dispute is whether the accused algorithm's "updating" step functions in the specific manner claimed. Practitioners may focus on this term because many modern optimization algorithms use ensemble methods or weighted averages of good performers, which may differ from the patent's more "elitist" approach of relying on the single best-performing vector.
- Intrinsic Evidence for Interpretation:
- Evidence for a Narrower Interpretation: The patent's own formula for the update step (v') shows it is a weighted sum of multiple sample vectors (X_i) that meet a performance threshold, not just the single best vector (E'). However, the weighting function itself is dependent on the best vector (E'), and the selection of which vectors to include in the sum is also based on their performance relative to a threshold derived from the best performers. (’120 Patent, col. 5:10-15, formula 2). This complex relationship may support an argument that the accused method, if different, does not meet the claim.
- Evidence for a Broader Interpretation: A party could argue that any update algorithm where the "best result" vector is given substantial weight or is used to define the bounds of the next search "uses" that vector in a manner consistent with the claim, even if it is not the sole input. The phrase "the properties of the best performing hyperparameters are used to guide the next iteration" could support this broader view. (’120 Patent, col. 2:25-27).
VI. Other Allegations
Indirect Infringement: The complaint seeks a declaration of non-infringement for both induced and contributory infringement. It alleges that because there is no underlying direct infringement of any claim, there can be no indirect infringement. It further asserts a lack of the requisite specific intent for inducement and a lack of specific knowledge for contributory infringement. (Compl. ¶¶ 39, 47).
VII. Analyst’s Conclusion: Key Questions for the Case
- A threshold procedural question will be one of justiciability: Did Defendant’s licensing communications and the provision of a draft amended complaint in a separate lawsuit create a sufficiently immediate and real controversy to grant Plaintiff standing to bring this declaratory judgment action?
- A core technical question for the ’717 patent will be one of routing architecture: Do the Accused Thread Products establish network routes by building directly upon the "previously established" route of the granting station as claimed, or do they employ a different, more dynamic mesh routing protocol that falls outside the claim's specific hierarchical requirements?
- For the ’120 patent, the key dispute will be one of algorithmic mechanism: Does the "updating" step in Google's Vertex AI platform rely on the single "hyperparameter vector producing the best result" from all prior iterations to calculate the next estimate, or does its algorithm use a different methodology, such as a weighted average of multiple high-performing vectors, that is technically distinct from the claimed method?