17937204. DRIFT MODE ACQUISITION USING SUCCESSIVE MODEL TRAINING simplified abstract (Dell Products L.P.)

DRIFT MODE ACQUISITION USING SUCCESSIVE MODEL TRAINING

Organization Name

Dell Products L.P.

Inventor(s)

[[:Category:Herberth Birck Fr�hlich of Florianópolis (BR)|Herberth Birck Fr�hlich of Florianópolis (BR)]][[Category:Herberth Birck Fr�hlich of Florianópolis (BR)]]

Vinicius Michel Gottin of Rio de Janeiro (BR)

DRIFT MODE ACQUISITION USING SUCCESSIVE MODEL TRAINING - A simplified explanation of the abstract

This abstract first appeared for US patent application 17937204 titled 'DRIFT MODE ACQUISITION USING SUCCESSIVE MODEL TRAINING

Simplified Explanation

The abstract describes a method for evaluating the drift mode of a machine learning model by comparing inference values generated by a reduced reference model with new versions of the same model over a defined time window.

• Define a time window for evaluating drift mode starting at time t.
• Train a reduced reference model R with a subset of the training dataset.
• Iteratively train new versions of R with new data samples.
• Collect data and compare inference values between R and R.
• Determine drift mode based on the drift mode curve generated from the absolute differences.

Potential Applications

This technology can be applied in various industries where machine learning models are used, such as finance, healthcare, and manufacturing, to monitor and detect drift in model performance over time.

Problems Solved

This technology addresses the challenge of detecting drift in machine learning models, which can lead to inaccurate predictions and decisions if not monitored and corrected in a timely manner.

Benefits

- Improved model performance monitoring - Early detection of drift in machine learning models - Enhanced decision-making based on accurate model predictions

Potential Commercial Applications

"Drift Mode Evaluation Method for Machine Learning Models" can be utilized in industries such as financial services for fraud detection, healthcare for patient diagnosis, and manufacturing for quality control to ensure the reliability and accuracy of machine learning models.

Possible Prior Art

One possible prior art could be methods for model evaluation and monitoring in machine learning, such as techniques for detecting concept drift or model degradation over time.

How does this method handle concept drift in streaming data scenarios?

This article does not address how the method adapts to concept drift in real-time data streams where the underlying data distribution may change over time.

What are the computational requirements for implementing this method at scale?

The article does not provide information on the computational resources needed to train multiple versions of the reduced reference model and compare inference values over a large dataset.

Original Abstract Submitted

One example method includes defining a time window during which a drift mode of a machine learning model will be evaluated, and the time window begins at t, training, beginning at the time t, a reduced reference model Rwith a data sample dthat is a subset of a training dataset that was used to train the machine learning model; for ‘n’ iterations: at a time t, when new data samples dare available, train a new version Rof the reduced reference model R; after Ris trained, collect data v; and compare respective inference values generated by the reduced reference model Rand R, using the data samples dand v, and store an absolute difference between the inference values; defining a drift mode curve using the absolute differences; and based on the drift mode curve, determining a drift mode of the machine learning model.