5:24-cv-00967
Andrew Kamal v. Femtosense Inc
I. Executive Summary and Procedural Information
Parties & Counsel:
- Plaintiff: Andrew Kamal (Michigan)
- Defendant: Femtosense, Inc. (California) and Sam Fok (California)
- Plaintiff’s Counsel: Andrew Kamal (pro se)
Case Identification: 5:24-cv-00967, C.D. Cal., 05/06/2024
Venue Allegations: Venue is alleged to be proper in the Central District of California on the basis that the Defendants reside in San Bruno, California.
Core Dispute: Plaintiff alleges that Defendants’ method for "sparsing conventional networks" infringes a patent related to a two-phase data compression method that preserves complex data types.
Technical Context: The technology concerns data compression algorithms, particularly for large scientific datasets where non-standard values like irrational or complex numbers are analytically significant.
Key Procedural History: The complaint attaches a pre-suit notice letter, dated April 15, 2024, which provides the substantive basis for the infringement allegations and establishes pre-suit knowledge of the patent.
Case Timeline
| Date | Event |
|---|---|
| 2018-08-09 | '315 Patent Priority Date |
| 2021-03-30 | '315 Patent Issue Date |
| 2024-04-15 | Pre-suit notice letter sent |
| 2024-05-06 | Complaint Filing Date |
II. Technology and Patent(s)-in-Suit Analysis
U.S. Patent No. 10,965,315 - "Data Compression Method," issued March 30, 2021
The Invention Explained
- Problem Addressed: The patent addresses a problem in data analytics where conventional software performs "data scrubbing," a process that omits or discards unexpected data such as irrational or complex numbers (e.g., numbers in the form a+bi) ('315 Patent, col. 1:19-25). This is described as problematic for analyzing certain scientific data, such as from particle accelerators or cancer genomics, where these "anomalous" numbers may represent the most significant data of interest ('315 Patent, col. 1:26-32).
- The Patented Solution: The invention proposes a two-phase method for compressing a data set while preserving this significant data. In a first "training" phase, the system analyzes an initial portion of a data set, sorting values based on predefined criteria into either a "first category" of "complexities" (which are kept) or a "second category" of "memoryless" data (which are excluded) ('315 Patent, col. 5:15-20). The excluded "memoryless" data is used to build a statistical distribution model ('315 Patent, col. 5:18-23). In a subsequent second phase, the system uses this statistical model, rather than the original complex criteria, to more efficiently categorize the remaining data ('315 Patent, col. 5:24-33). This process is illustrated in the flowchart of Figures 3A-3B.
- Technical Importance: This approach is presented as a method to efficiently compress very large datasets while ensuring that critical, non-standard data points ("complexities") are not scrubbed, thereby improving the integrity of subsequent analysis ('315 Patent, col. 5:57-61).
Key Claims at a Glance
- The complaint asserts independent claim 1 and dependent claims 2-13 and 18-21 (Compl. ¶9, p. 20).
- Independent Claim 1 requires:
- obtaining a data set and criteria for determining whether values correspond to a first or second category
- determining that some values correspond to the first category and others to the second
- adding values from the first category to a compressed data set
- excluding values from the second category from the compressed set, and using them to update a statistical distribution
- performing this determination during a first phase based on comparison to the criteria
- performing this determination during a second phase based on the statistical distribution
III. The Accused Instrumentality
Product Identification
The accused instrumentality is identified as Femtosense, Inc.'s "method for sparsing conventional networks" (Compl. ¶9, p. 19).
Functionality and Market Context
The complaint alleges that Femtosense applies a form of compression in its "electronics manufacturing" that involves a "quantization method" (Compl. ¶9, p. 20). This method is alleged to utilize "Bayesian learning in relation to weights and polarity," a "Quadtree data structure," and applies "different weights for the number or data types" (Compl. ¶9, p. 20). The complaint references an alleged statement by Femtosense, "weights that matter," to support the allegation that the accused method differentiates data based on importance, similar to the patent's categories (Compl. ¶9, p. 20). The complaint further alleges that Femtosense's quantization representation resembles a predecessor to the Quadtree structure shown in the patent's Figure 5 (Compl. ¶9, p. 20). A figure in the patent shows a schematic view of an example point quadtree data structure, which divides a data space into four quadrants (Q1-Q4) for hierarchical data storage ('315 Patent, Fig. 5).
IV. Analysis of Infringement Allegations
'315 Patent Infringement Allegations
| Claim Element (from Independent Claim 1) | Alleged Infringing Functionality | Complaint Citation | Patent Citation |
|---|---|---|---|
| obtaining a data set and criteria for determining whether individual values from the data set correspond to a first category or a second category of values | The accused method allegedly uses criteria such as "different weights for the number or data types" and "complexities for zeros" to classify data. The patent's Figure 3A is cited as showing a parallel process. | ¶9 | col. 1:35-39 |
| determining that some values of the data set correspond to the first category, and that other values of the data set correspond to the second category | The accused method allegedly distinguishes between data types to determine which "weights... matter," which is alleged to correspond to the patent's categorization. | ¶9 | col. 1:39-41 |
| based on one of the values corresponding to the first category, adding the value to a compressed data set | Femtosense is alleged to apply a "similar form of compression to their electronics manufacturing," which implies that certain data is retained in a compressed form. | ¶9 | col. 1:42-43 |
| based on one of the values corresponding to the second category: excluding the value from the compressed data set; and updating a statistical distribution... based on the value | The alleged use of "Bayesian learning," "neural learning," and "recursion" is asserted to meet this limitation, suggesting a statistical model is built and updated based on excluded or processed data. | ¶9 | col. 1:44-46 |
| wherein during a first phase, the determining is performed for a plurality of values... based on comparison of the values to the criteria | The allegations point to "specialized instructions, tuning, and optimizing" as evidence of an initial phase where data is processed against set rules, analogous to the patent's first phase. | ¶9 | col. 1:46-50 |
| wherein during a second phase that is subsequent to the first phase, the determining is performed... based on the statistical distribution | The alleged use of "Bayesian learning" and "neural learning" is asserted to be the second phase, where decisions are made based on a learned model rather than the initial, fixed criteria. | ¶9 | col. 1:50-54 |
Identified Points of Contention
- Scope Questions: A central question will be whether Femtosense's "method for sparsing conventional networks" can be properly characterized as a "method of compressing a data set" as recited in the claim preamble.
- Technical Questions: The complaint alleges that Femtosense's method is a two-phase process. A key factual question is whether the accused method actually contains a first phase that relies exclusively on pre-set criteria, followed by a distinct second phase that relies exclusively on a statistical model built during the first, as the claim requires.
- Technical Questions: What evidence demonstrates that Femtosense's alleged use of "Bayesian learning" functions as the claimed "statistical distribution" updated by "memoryless" data, and is used for data categorization in a second phase?
V. Key Claim Terms for Construction
The Term: "first category" / "second category"
- Context and Importance: The infringement theory hinges on mapping Femtosense's alleged data classification (e.g., "weights that matter") onto the patent's specific two-category system. The outcome of the case may depend on whether these terms are construed broadly to cover any binary classification or narrowly to the specific types of data described in the patent.
- Intrinsic Evidence for a Broader Interpretation: The claim language itself does not define the categories, stating only that they are determined by "criteria," which could support a construction covering any rule-based, binary sorting of data (e.g., important vs. unimportant).
- Intrinsic Evidence for a Narrower Interpretation: The specification repeatedly defines the "first category" as "complexities" and provides specific examples, including irrational numbers, complex numbers, and mixed hashes ('315 Patent, col. 6:7-18). The "second category" is defined as "memoryless data," such as integers and zeros ('315 Patent, col. 5:18-19). This may support a narrower construction limited to this specific technical division.
The Term: "statistical distribution"
- Context and Importance: This term is critical to the "second phase" of the claimed method. The infringement allegation relies on equating Femtosense's alleged "Bayesian learning" with this claim element. The construction of this term will determine what level of technical proof is required to show infringement of the second phase.
- Intrinsic Evidence for a Broader Interpretation: The term itself is general and could be argued to encompass any statistical model derived from input data for the purpose of classification.
- Intrinsic Evidence for a Narrower Interpretation: The specification describes the statistical distribution as being composed of "memoryless data" and indicates "how often those values appear in the data set" ('315 Patent, col. 5:21-23). The patent further describes using Bayes' theorem with this distribution to calculate the probability that a value is "memoryless" ('315 Patent, col. 8:1-8). This may support a construction requiring a model that specifically tracks the frequency of excluded data points.
VI. Other Allegations
Indirect Infringement
The complaint's prayer for relief seeks findings of induced and contributory infringement (Compl. p. 2-3). The complaint establishes pre-suit knowledge via the attached notice letter (Compl. ¶9). However, the complaint does not provide sufficient detail for analysis of indirect infringement, as it does not allege specific facts regarding how Defendants might encourage or provide components for infringement by a third party.
Willful Infringement
The complaint does not contain a formal count for willful infringement. However, it alleges that Defendants had pre-suit notice of the patent and their alleged infringement via a letter dated April 15, 2024, and "continued to infringe" thereafter (Compl. ¶9). The letter itself alleges "deliberate and intentional infringement" (Compl. p. 19). These allegations may form the basis for a future claim of post-filing willfulness.
VII. Analyst’s Conclusion: Key Questions for the Case
- A core issue will be one of technical mapping: Can the plaintiff produce evidence to demonstrate that the accused "method for sparsing conventional networks" is in fact a two-phase algorithm that mirrors the patent's structure—a first phase using fixed criteria to build a statistical model, followed by a second phase that uses only that model for classification?
- A key evidentiary question will be one of functional equivalence: Does the alleged use of "Bayesian learning" and "weights" in the accused method perform the same function, in the same way, to achieve the same result as the claimed "statistical distribution" that is specifically built from "memoryless" data to categorize subsequent data points?
- The case may also turn on a definitional question: Can the patent’s claim terms "first category" and "second category" be construed broadly enough to read on Femtosense's method of data prioritization, or will they be limited by the specification to the specific division between "complexities" (e.g., irrational numbers) and "memoryless" data (e.g., integers)?