This inventor holds 2 USPTO granted patents and 1 published patent application. Top assignee: Five Ai Limited. Active years: 2026.
Data source: USPTO (United States Patent and Trademark Office) public patent records. Weekly synchronization. How IDiyas builds this profile
Company Filing History:
Years Active: 2026
Title: Puneet Dokania: Inventor in Sensor-Data Feature Extraction.
Introduction
Puneet Dokania is an inventor based in Cambridge, Great Britain. His work includes patent inventions focused on extracting useful features from sensor data for machine-learning and perception systems. He is associated with Five AI Limited and has 2 patents.
Latest Patents
Dokania’s latest patent work addresses methods for training encoders to extract features from sensor data.
One invention describes alternating between two training phases. In the first phase, an encoder learns from unannotated sensor data by optimizing a self-supervised loss function, while the perception component’s task-specific parameters remain unchanged. In the second phase, annotated sensor data is processed by the updated encoder and perception component, and the task-specific parameters are adjusted to optimize performance against the annotations. Repeating these phases enables the system to learn representations from unannotated data while also training for a desired perception task.
A related invention trains an encoder using pairs of data representations derived from a common set of sensor data. A self-supervised loss encourages the machine-learning system to associate each first representation with its corresponding second representation based on their extracted features. To generate the second representation, the method applies a 2D object detector to an associated image and transforms the sensor data according to the detected objects. This approach uses visual detection to guide the creation of training data for sensor-based feature learning.
Career Highlights
Dokania’s listed patent portfolio comprises 2 patents, with both described inventions concerning feature extraction from sensor data. His work combines self-supervised learning with object detection and task-specific perception training.
Collaborations
Dokania is associated with Five AI Limited. His listed coworkers include John Redford and Anuj Sharma.
Conclusion
Puneet Dokania’s patent work focuses on training machine-learning encoders to extract informative features from sensor data, supporting perception systems through self-supervised and task-specific learning methods.
Data source: USPTO (United States Patent and Trademark Office) public patent records. Weekly synchronization. How IDiyas builds this profile