DCT

1:26-cv-00754

Kmizra LLC v. Google LLC

Log In
Key Events
Amended Complaint
complaint Intelligence

I. Executive Summary and Procedural Information

  • Parties & Counsel:
  • Case Identification: 1:26-cv-00754, W.D. Tex., 09/23/2026 (Filing date of the Notice of First Amended Complaint)
  • Venue Allegations: The provided document does not detail the basis for venue allegations.
  • Core Dispute: The complaint detailing the specific allegations was not provided; however, based on the identified patents, the dispute likely concerns allegations that Defendant's products or services infringe patents related to wireless network initialization and machine learning hyperparameter optimization.
  • Technical Context: The technologies at issue relate to foundational aspects of modern computing: the formation of ad-hoc wireless networks and the automated tuning of machine learning models.
  • Key Procedural History: The provided notice indicates that Plaintiff filed a First Amended Complaint (ECF No. 32) in response to Defendant’s Corrected Motion to Dismiss (ECF No. 30), rendering the motion moot.

Case Timeline

Date Event
2005-12-30 ’717 Patent Priority Date
2007-04-25 ’120 Patent Priority Date
2012-03-27 ’717 Patent Issued
2013-05-07 ’120 Patent Issued
2026-09-23 Notice of First Amended Complaint Filed

II. Technology and Patent(s)-in-Suit Analysis

U.S. Patent No. 8,144,717 - Initialization of a wireless communication network, issued March 27, 2012 (’717 Patent)

The Invention Explained

  • Problem Addressed: The patent describes the inefficiency and complexity of establishing routing topologies in wireless sensor networks. Prior art methods required forwarding information through stations whose own routes to a central point were not yet defined, creating a complicated and potentially slow setup process ’717 Patent, col. 1:59 - col. 2:3
  • The Patented Solution: The invention proposes a structured, wave-like association process that radiates outward from a central "association unit." New stations begin in a "not-associated state" and send out "association request messages." Only the association unit or stations that are already in an "associated state" can issue an "association grant message." Upon receiving a grant, a station switches to the "associated state" and establishes a defined route back to the central unit through the granting station. This ensures that the network is built upon a foundation of established, reliable links ’717 Patent, abstract ’717 Patent, col. 3:15-38 Figure 1 illustrates this topology, with a central unit (10) connecting to beacon frame transmitters (12), which in turn connect to sensor stations (14).
  • Technical Importance: This method offers a systematic approach to initializing ad-hoc or mesh networks, which can improve the speed, reliability, and power efficiency of network formation compared to less structured protocols.

Key Claims at a Glance

  • The complaint was not provided, but Claim 1 (a network claim), Claim 11 (a station claim), and Claim 14 (a method claim) are the independent claims. Analysis will focus on Claim 1.
  • Independent Claim 1 requires:
    • An association unit and a plurality of stations, with each station configured to start in a "not-associated state."
    • Stations transmit "association request messages" while in the not-associated state.
    • Stations switch to an "associated state" upon receiving an "association grant" in response to a request.
    • The grant establishes an "operating route associated with the station" that runs through the source of the grant.
    • The association unit is configured to transmit grants.
    • At least some stations are configured to transmit grants, but "only after switching to the associated state."

U.S. Patent No. 8,438,120 - Machine learning hyperparameter estimation, issued May 7, 2013 (’120 Patent)

The Invention Explained

  • Problem Addressed: The patent addresses the difficulty of finding optimal hyperparameters for machine learning classifiers. The search space is often extremely large, making simple grid searches ineffective, while more sophisticated statistical methods like the cross-entropy (CE) method are not considered suitable for this specific problem ’120 Patent, col. 1:49-59
  • The Patented Solution: The invention discloses an iterative optimization method. In each iteration, the system generates a random sample of hyperparameter vectors and evaluates their performance. The key step is selecting the hyperparameter vector that has produced the "best result" in the current or any previous iteration. This "elitist" or "best-so-far" vector is then used to update the target estimate, guiding the random sampling in subsequent iterations toward more promising regions of the search space ’120 Patent, abstract ’120 Patent, col. 2:2-19 ’120 Patent, col. 5:27-34
  • Technical Importance: This method provides a more efficient, guided approach to hyperparameter tuning than exhaustive or purely random searches, a critical process for deploying high-performance machine learning models.

Key Claims at a Glance

  • The complaint was not provided, but Claim 1 (a method claim) and Claim 14 (a device claim) are the independent claims. Analysis will focus on Claim 1.
  • Independent Claim 1 requires:
    • A method of determining hyperparameters of a classifier by iteratively producing an estimate of a target hyperparameter vector.
    • Each iteration comprises the steps of:
      • "drawing a random sample of hyperparameter vectors."
      • "updating the estimate of the target hyperparameter vector by using the random sample."
      • "selecting, from the random sample of hyperparameter vectors, a hyperparameter vector producing a best result in the present and any previous iterations."
    • The updating step "uses said hyperparameter vector producing the best result."

III. The Accused Instrumentality

The provided document, a notice of an amended complaint, does not identify the accused product(s), method(s), or service(s). Therefore, no analysis of the accused instrumentality is possible.

No probative visual evidence provided in complaint.

IV. Analysis of Infringement Allegations

The complaint containing the specific infringement allegations was not provided. As a result, an analysis of the infringement allegations, including the creation of a claim chart, cannot be performed.

V. Key Claim Terms for Construction

  • Term from the ’717 Patent: "associated state"

    • Context and Importance: The distinction between a "not-associated state" and an "associated state" is fundamental to the claimed invention's sequential, hierarchical network formation process. Infringement will depend on whether an accused device can be shown to operate in these two distinct states as defined by the patent. Practitioners may focus on this term to determine if an accused system's network participation status maps onto the claimed binary states.
    • Intrinsic Evidence for Interpretation:
      • Evidence for a Broader Interpretation: The patent does not appear to mandate a specific software flag or register for the state. A party could argue that any device that is actively routing traffic within a network, regardless of the underlying protocol, is functionally in an "associated state".
      • Evidence for a Narrower Interpretation: The specification links the "associated state" to having a "well defined route" to the association unit, established in response to a specific grant message ’717 Patent, col. 10:62-64 ’717 Patent, col. 13:1-4 A party may argue this requires a specific, recorded routing path established via the claimed request/grant mechanism, not just general network connectivity.

  • Term from the ’120 Patent: "best result"

    • Context and Importance: This term is the linchpin of the selection step that drives the optimization process. The definition of "best" is critical to determining whether an accused system performs the claimed selection. A dispute could arise over what performance metric(s) constitute a "best result" and how that result is determined.
    • Intrinsic Evidence for Interpretation:
      • Evidence for a Broader Interpretation: The specification provides flexibility, stating the function S(X) used to find the best vector "may be a loss function or any other suitable function producing a result value based upon X" ’120 Patent, col. 5:29-31 This language suggests that any quantifiable performance metric (e.g., accuracy, precision, F1-score, inference speed) could qualify.
      • Evidence for a Narrower Interpretation: While the patent is broadly written, a party might argue that the context implies a single, scalar "result value" that can be unambiguously ranked (’120 Patent, col. 5:35-39, describing ranking of S(X)` values). An accused system using multi-objective optimization, where there may be a set of non-dominated "best" solutions (a Pareto front) rather than a single best one, could raise the question of whether it meets this limitation.

VI. Other Allegations

The complaint was not provided. Therefore, no analysis of any allegations of indirect infringement or willful infringement is possible.

VII. Analyst’s Conclusion: Key Questions for the Case

As the complaint was not available for review, the central questions are framed based on the patent claims and the likely nature of potential accused products from a large technology company.

  1. ’717 Patent - A Question of Structural Correspondence: A core issue for the ’717 Patent will be whether the dynamic and often decentralized networking protocols used in modern wireless systems (e.g., mesh Wi-Fi, peer-to-peer discovery) perform the specific, sequential steps required by the claims. The key question is one of structural correspondence: does an accused system that forms a network utilize the claimed sequence of "not-associated" to "associated" states, driven by explicit "association request" and "association grant" messages, where only already-associated nodes can propagate the network?

  2. ’120 Patent - A Question of Algorithmic Uniqueness: For the ’120 Patent, the dispute will likely center on the precise mechanics of the hyperparameter optimization algorithm. The central question is one of algorithmic uniqueness: does an accused hyperparameter tuning system (such as Google's Vertex AI Vizier) implement the specific "elitist" method of selecting a vector with the "best result" from the present and all previous iterations to guide the next sampling, or does it employ a distinct optimization strategy (e.g., pure Bayesian optimization, population-based methods) that does not rely on this specific claimed feedback mechanism?