Questions about how individual parts of claim 1 might be changed. A question may describe something different from what the claim language recites, or a version of it; no view is offered on which.
1.2Removable checkpoint storage card
What ifWhat if a removable storage card carried checkpoint records between game units that lack network interface controllers?
Claim language“at least one network interface controller”
Each unit uses a storage controller to write a checkpoint identifier, a sequence number, and a checksum to the card. When the card moves to another unit, that unit verifies the checksum and loads the scene identified by the latest intact record.
1.3Fixed checkpoint rules
What ifWhat if fixed rules selected the restart location without a trained model?
Claim language“a trained machine-learning model”
A rule table assigns a checkpoint to each combination of current zone and completed objective flags. The game selects the most specific matching row and uses the last recorded checkpoint when no row matches.
1.3Forward-only checkpoint sequence
What ifWhat if restart selection operated on a checkpoint sequence that allowed movement only toward later checkpoints?
Claim language“a bi-directional game timeline”
The save manager stores one active checkpoint index and accepts an update only when all required event flags for the next checkpoint are set. On launch, the game loads the scene at the stored index and continues from that point.
1.3Restart state compatibility checks
What ifWhat if the game accepted a predicted restart location based on state compatibility without checking prediction confidence?
A validator checks whether the scene assets are present and whether the saved state contains the event flags and inventory entries required by the scene. A passing result allows the profile to store the predicted location, and a failing result retains the previous location.
Section 2 · machine-generated
Variations
Further arrangements of the same subject matter, described here as of the publication date at the foot of this page.
Other field of use
Restart prediction for maintenance training
A networked maintenance simulator records trainee movement between work areas, tool selections, and time spent at each service station. A trained classifier scores restart markers in a timeline that supports forward and backward navigation through the simulated procedure. The session server stores the highest-scoring marker in the trainee profile when its score meets the configured confidence threshold. On return, the simulator restores the equipment configuration and trainee tool state from a snapshot and the state changes up to that marker.
Component substitution
Nearest neighbor restart location prediction
A weighted k-nearest neighbors classifier replaces the restart predictor and stores labeled session vectors with restart marker identifiers. The timeline controller scales movement, inventory activity, and zone dwell features using the means and standard deviations computed from the training records. It finds the 15 nearest vectors by Euclidean distance and selects the marker with the largest inverse-distance vote total, using a small positive offset in the distance denominator. The controller divides the winning vote total by the total weight of all 15 votes and writes the marker to the player profile when that fraction meets the confidence threshold.
Parameter or scale variation
Fixed windows for session scoring
A game server records player position twice per second and logs inventory transfers and zone transitions as they occur. At session end, the timeline controller aggregates up to 120 seconds of data into twelve 10-second windows and attaches a presence mask for windows with missing data. A trained model scores up to 24 restart markers drawn from the 10 minutes before and after the exit position on the timeline. The controller writes the highest-scoring marker to the player profile when its calibrated probability is at least 0.85.
Combination with existing systems
Restart association with cloud save synchronization
A timeline service integrates with a cloud save system that assigns a revision identifier to each synchronized save. The trained model scores restart markers that the engine can reconstruct from the current save revision and its stored state changes. When the winning score meets the confidence threshold, the service stores the marker and save revision together in the player profile using a conditional write that checks the current revision. On another device, the game retrieves the referenced revision and applies its state changes through the stored marker before enabling player input.
Automated or data-driven variant
Supervised updates from player marker selections
The server records the session feature vector and predicted marker for each automatic restart. When a player selects a different marker with the timeline scrubber within 30 seconds and plays from it for at least two minutes, a labeling service pairs that marker with the original feature vector. A scheduled job retrains the classifier on these pairs, calibrates its scores on a separate validation set, and publishes the model and calibration data together. The timeline controller uses the published model to score subsequent sessions and updates a profile when the selected marker meets the confidence threshold.
Restart selection with mandatory event checks
The game engine stores an event completion record with each timeline checkpoint and lists the mandatory event identifiers required at each restart marker. The timeline controller filters markers by checking whether each marker's reconstructed state contains its required completion records. A trained model uses player movement, inventory activity, and zone dwell time to score the eligible markers on both sides of the current timeline position. The controller stores the highest-scoring eligible marker in the player profile when its confidence meets the configured threshold.
Specific version of a part
Hardwired checkpoint selection circuit
A priority encoder selects the highest eligible checkpoint from a 32-bit register that records completed events. On an exit pulse, a fixed state machine writes the selected index to nonvolatile memory and restores it to an output register at power-up.
Four-core session processor
A four-core processor executes session logging and restart selection. Worker threads process records for separate players, and a per-profile mutex serializes concurrent updates to the same restart location.
Gigabit wired network controller
A network interface controller connects to a local switch through a twisted-pair cable at one gigabit per second. Direct memory access transfers received session packets into buffers in system memory.
Checkpoint classifier with confidence gating
Processor-addressable memory holds timeline control instructions and a gradient-boosted tree classifier trained on session records labeled with accepted restart locations. The instructions pass zone dwell times and inventory decision counts to the classifier, which scores candidate points on a timeline that supports movement in both directions. The instructions update the profile with the highest-scoring point when its predicted probability reaches 0.85.
Section 3 · quoted from USPTO records
Other work
Passages reproduced word for word from two earlier-published U.S. patent documents, placed beside the language of claim 1. No characterization is offered. Each passage is marked A or B for the document it comes from; ¶ numbers are the paragraph numbers in the USPTO’s published full text of that document.
A microprocessor used in a pair with a baseband processor for performing the baseband processing, is provided with a central processing unit for calculation processing, a counter capable of measuring time in the calculation processing by the central processing unit, and an interface which enables the baseband processor to read the counter.
A · US 2005/0012811 A1 · abstract
1.2at least one network interface controller configured to provide access to a network; and
1.3a memory communicatively coupled to the processor, wherein the memory comprises a timeline control logic that is configured to: collect player session data comprising at least one of player movement patterns, inventory management decisions, or time spent in specific game zones; feed the player session data into a trained machine-learning model to generate an optimal restart location prediction within a bi-directional game timeline; determine if the optimal restart location prediction is within a pre-determined confidence threshold; and update a timeline association for a player profile to the optimal restart location prediction to facilitate automatic progression management.
An electronic device comprising: at least one processor; and at least one memory operatively coupled to the at least one processor to store at least one instruction for causing the at least one processor to perform operations, wherein the operations include performing a learning operation of an artificial intelligence model that predicts a label with respect to input data stored in the at least one memory, wherein the learning operation includes obtaining a prediction confide
B · US 2022/0019891 A1 · claim 1
An empty cell means only that no passage was reproduced there.
Statements in these documents
This, however, requires adjustment of reproducing timing between video and its corresponding voice and some device for establishing synchronization between the video processed in the application processor and the voice processed in the baseband processor.
A · US 2005/0012811 A1 · ¶ 7
An object of the invention is to provide a technique for establishing synchronization between video and voice when the video processing and the voice processing are separately performed by the different processors.
A · US 2005/0012811 A1 · ¶ 8
Conventional DNN learning requires a very large amount of computation, so there is a problem that it consumes a long learning time and a lot of learning energy.
B · US 2022/0019891 A1 · ¶ 4
Embodiments of the present disclosure provide an electronic device and a method for learning of a low-complexity artificial intelligence model based on dynamic prediction confidence threshold selection in which a user is able to adjust the degree of omission of the amount of a change in weight through dynamic prediction confidence threshold selection.
B · US 2022/0019891 A1 · ¶ 6
Cite aspatent.fyi, US 2026/0355639 A1, claim 1, first published 2026-11-19T13:37:15-05:00, https://patent.fyi/US20260355639/, SHA-256 e8230c865502600ab04b725be0f9eb440aada0b4a5e39500201c681fe2db3ca1