Continuous inference for sequence data
View Patent ↗Dynamic inference is leveraged to provide online sequence data labeling. This provides real-time alternatives to current methods of inference for sequence data. Instances estimate an amount of uncertainty in a prediction of labels of sequence data and then dynamically predict a label when an uncertainty in the prediction is deemed acceptable. The techniques utilized to determine when the label can be generated are tunable and can be personalized for a given user and/or a system. Employed decoding techniques can be dynamically adjusted to tradeoff system resources for accuracy. This allows for fine tuning of a system based on available system resources. Instances also allow for online inference because the inference does not require knowledge of a complete set of sequence data.
1. A system for providing continuous inference for sequence data, comprising a processor, wherein the processor comprises the following executable instructions:
an estimation component that estimates an amount of uncertainty in a prediction of a label for a sequence data observation, the estimation component utilizes subsequent sequence data observations to re-evaluate the amount of uncertainty as each observation is received;
a delay component that monitors an amount of delay since receiving an initial sequence data observation; and
a determination component that employs an algorithm that dynamically selects and outputs a label for the sequence data observations when the amount of uncertainty and the amount of delay is within a threshold, the threshold determined by finding an output time t that minimizes the function M(t,t 0 |o)+λ·(t−t 0 ), where t 0 is the time of a first sequence data observation, M(t,t 0 |o) is a measure of residual uncertainty in the label at time step t 0 given observations o, and λ is a tunable parameter that indicates a preferred balance between label output delay and label confidence.
2. The system of claim 1 further comprising:
a parameter component that determines the tunable parameter based on at least one of a user preference or available system resources.
3. The system of claim 2 , the available system resources including one or more of battery power, memory, or processor.
4. The system of claim 1 , the determination component yields to an absolute delay constraint that forces a label selection regardless of the threshold after an amount of delay equal to the delay constraint has elapsed.
5. The system of claim 4 , the absolute delay constraint is based on the tunable parameter.
6. The system of claim 1 , the determination component selects the label using the tunable parameter based on a tradeoff between accuracy and at least one of the amount of delay, memory usage, or processing usage.
7. The system of claim 1 , the tunable parameter is influenced by accuracy constraints from one or more system resources.
8. The system of claim 7 , the tunable parameter continuously adjusts accuracy constraints based on available battery power of a mobile device.