SigOpt is an optimization service that helps customers find their best machine learning models. Our simple API makes suggestions to customers about different versions of their model to try, they evaluate those suggestions, and report back to us how well they did, so we can build our own predictions of what's going to make their model perform the best.

Product

• Optimization Engine intelligently finds model parameter suggestions using techniques like Bayesian optimization.
• Simple API enables data scientists and machine learning practitioners to connect to our service within their existing machine learning pipelines
• Experiment Insights web dashboard empowers researchers to introspect and reproduce experiments through the model development process.

Team

Our team is built from world experts in optimization, leaders from the biggest names in tech, talented engineers, and ML practitioners. Together, we're unlocking the promise of optimization for everyone.

Example Projects

Orchestrate: Enable our customers to find their optimal model faster by embedding their training process into a docker image, and trying different parameter configurations in parallel in a kubernetes cluster.

Multimetric: Optimize two metrics simultaneously, for example the accuracy vs size of a deep learning model. Involves both research into how to sample a more complex space, and practical considerations of how to make visualizations and an API that customers can intuitively use.

Teams and Projects: Securely and easily group users or experiments together to share insights about experiments. Develop a permissioning model, APIs, and UI for a cohesive product release.

On Premises High Availability: Develop a set of Docker images that customers can deploy to their private cloud that minimize outages and provide a process to easily upgrade versions.

Deep Learning Experiments: Learn from intermediate results while training a deep learning model so that poor performing hyperparameters can be avoided, helping customers find optimal models for expensive models.

Values

• Empowerment: Take ownership and make bold decisions
• Curiosity: Be a humble teacher and active student
• Respect: Foster an inclusive, professional and safe environment for everyone
• Balance: Know when to work, when to play and when to go home
• Solidarity: Transparently collaborate toward shared goals

If you're interested in making the world a better place with the power of software and algorithms, reach out to learn more.