Tune Your ML Models to Drive Performance and Revenue

DZone 's Guide to

Tune Your ML Models to Drive Performance and Revenue

When optimally tuned, your machine learning models can assist you in quickly going from trial-and-error to improved results.

· AI Zone ·
Free Resource

It was great talking with Scott Clark, CEO and Co-founder of SigOpt. Scott founded SigOpt nearly three years ago after doing his Ph.D. research in applied math at Cornell and seeing how much time was required to fine-tune deep learning models.

Q: How are you and your company involved in machine learning (ML)?

A: SigOpt automates the tuning of ML models’ hyper, feature, and architecture parameters to improve performance and drive revenue gains.

Q: What are the keys to a successful ML strategy?

A: Everything companies do should be tied to the business value they provide. ML is great at solving problems, but ML models and parameters must be fine-tuned to enable those models to arrive at the best possible outcomes. Organizations should have a contextual understanding of the business value they want ML to provide to maximize its potential.

Q: How can companies benefit from ML?

A: ML can do what humans cannot do or what humans aren’t very good at doing. For example, ML can identify fraudulent transactions or look at 100 variables at once and recognize a pattern. ML is able to pull a signal out from a lot of noise. In addition, ML is useful for image classification, algorithmic trading, and quickly optimizing trial and error to improve results.

Q: How has ML changed in the past year?

A: Users have gone from applying first generation machine learning — like decision trees and regression — to more sophisticated deep learning and artificial intelligence (AI). This is all thanks to the availability of infrastructures like GPUs in the cloud, as well as access to open-source software libraries like TensorFlow, MXNet, and Caffe2. They’re using more sophisticated tools and seeing more significant results.

Q: What are the technical solutions you and your clients are using for your ML initiatives?

A: TensorFlow, MXNet, and Caffe2 have displaced much of the older technology. We also see H2O, scikit-learn, and other open-source tools as popular solutions.

Q: What are some real-world problems you and your clients are solving with ML?

A: In financial services, algorithmic traders have the ability to make trades in a matter of milliseconds, using cutting edge tools like SigOpt to tweak peak performance. Banks and credit card companies also use SigOpt to optimize their risk management, fraud detection, and customer segmentation models. Hotwire is another SigOpt customer that uses ML to improve search ranking and VSA Partners is optimizing models for the ideal marketing mix. We’re seeing wide horizontal adoption across industries and all of them need help tuning their ML models to enhance performance.

Q: What are the most common issues you see preventing companies from realizing the benefits of ML?

A: Access to tools and infrastructure. Companies often struggle to fine tune the tools they download to optimize performance and resolve configuration issues. Another problem is structuring data so it can be ingested in a consumable format.

Q: Where do you think the biggest opportunities are in the implementation of ML?

A: There’s adoption across essentially every industry. The financial services and oil and gas industries have a history of using ML models they are looking to upgrade. Other industries like pharmaceuticals, manufacturing, and energy are starting to recognize the opportunity to leapfrog their competitors through ML solutions like SigOpt.

Q: What are your biggest concerns with ML today?

A: More companies need to make sure they are optimizing their processes. They should look to eliminate the most tedious part of data scientists’ jobs, so experts can focus on adding value rather than spending extensive time on tasks that ML can handle for them.

Q: What skills do developers need to be proficient on ML projects?

A: Developers need to learn as much as they can so they can be interdisciplinary and aren’t pigeonholed as a professional.  Developers can achieve this by becoming skilled in mathematics and statistics to help understand how to best apply ML models. Coursera is a great learning tool and I recommending developers look across its different zones to explore areas where they can have the broadest impact.

ai ,deep learning ,machine learning

Opinions expressed by DZone contributors are their own.

{{ parent.title || parent.header.title}}

{{ parent.tldr }}

{{ parent.urlSource.name }}