Deep Learning Applied to Sequence Data
Deep learning has shown great practical results in the area of application to complex sequence data. Patterson Consulting can help your organization better understand the ETL, vectorization, modeling, and deployment complexities inherent in leveraging deep learning methods for seqeunce data.
The presentation below is just one example how we took an open data set (PhysioNet Challenge 2012) and applied long short-term memory networks (LSTMs) to build a competitive ICU Mortality Prediction model.
Anomaly Detection
Many times companies want to know that "something strange happened" in their incoming data. This is typically referred to as "anomaly detection" and is commonly something we see in sequence datasets. From unbounded oscillations in phasor measurement unit data, to financial fraud in transaction data, to data that just "does not look normal", anomaly detection exists in most domains.
Patterson Consulting can help with deep experience in areas such as:
- Unbounded oscillation detection
- Unsupervised learning-based anomaly detection with autoencoders
- Complex timeseries vectorization methods (tensor creation, iSAX decomposition)
Sensor Data Ingest Strategies
Sequence data (sensor, timeseries, etc) tends to have special needs for collection at high throughputs and storage. Patterson Consulting can help your company in sizing your infrastructure for these challenges and making sure you have the right information architecture setup on the storage system. We help companies build sequence data collection systems that are well-suited for scalable ETL and vectorization pipelines.
Application Design + Implementation
Operationalize your investment by leveraging the Patterson Consulting team to deliver end-to-end transactional data application solutions to your organization.
- Identifying patterns in users on websites
- Predicting next-steps for shoppers
- Detect anomalous data as it arrives in real-time
- Generative text modeling
- Building supervised deep learning (LSTM) models for prediction
Sequence Data ETL and Vectorization
Sequence data tends to have specialized methods for ETL and vectorization depending on the specific use case. Some scenarios require custom Spark pipelines to join and order the data correctly. Our team has a long history in building these class of pipelines.
In the domain of applied machine and deep learning, we need to build tensor data structure from the sequence data ETL pipelines. The Patterson Consulting team can build out these specialized pipelines and help your organization leverage high-volume sequence data in their data science applications.
Sequence Data Modeling Feasibility Study
Many times customers wonder what is possible for a given custom model. We can spend some time with a customer's data to better understand the potential for a successful model project in the form of a feasibility study.
The Patterson Consulting team can work with you to determine the feasibility of the modeling project perform activities such as:
- Determining if there is enough data to model the desired outcome
- Determine if there is enough signal in the data to build a model
- Help the customer understand the bounds on how accurate the model can be (and if that is sufficient for their line of business use case).