Valohai Blog

Random hyperparameter optimization

[fa icon="calendar'] Dec 12, 2018 2:57:30 PM / by Aarni Koskela posted in New Features, Product Updates

Valohai now supports random search for hyperparameter optimization (which we call the Tasks feature), which has been proven in the aptly named paper Random search for hyper-parameter optimization to be an efficient way to find “neighborhoods” of likely-to-be-optimal hyperparameter values, which can then be iterated further to find the really good values.

Read more [fa icon="arrow-right"]

The Journey from Deep Learning Experimentation to Production-Ready Model Building

[fa icon="calendar'] Dec 3, 2018 11:10:57 AM / by Fredrik Rönnlund posted in Deep Learning in Production

Since the rise of the deep learning revolution, springboarded by the Krizhevsky et al. 2012 ImageNet victory, people have thought that data, processing power and data scientists were the three key ingredients to building AI solutions. The companies with the largest datasets, the most GPUs to train neural networks on, and the smartest data scientists were going to dominate forever.

Read more [fa icon="arrow-right"]

Watch the Webinar on Version Control in Machine Learning

[fa icon="calendar'] Nov 27, 2018 12:22:34 PM / by Joanna Purosto posted in Webinar, Version Control, Machine Learning

Watch a recording of the webinar on version control in machine learning that was held on 22th of November 2018. During the webinar we discussed about the topics below and answered multiple questions addressed by the attendees. 

Read more [fa icon="arrow-right"]

PocketFlow with Valohai

[fa icon="calendar'] Nov 26, 2018 2:18:00 AM / by Juha Kiili

PocketFlow is an open-source framework from Tencent to automatically compress and optimize deep learning models. Especially edge devices such as mobile phones or IoT devices can be very limited on computing resources so sacrificing a bit of model performance for a much smaller memory footprint and lower computational requirements is a smart tradeoff.

Read more [fa icon="arrow-right"]

Microsoft's Cognitive Toolkit (CNTK) on Valohai

[fa icon="calendar'] Nov 21, 2018 11:28:51 AM / by Eero Laaksonen posted in Microsoft, CNTK

Microsoft's Cognitive Toolkit or CNTK is an open source framework for building Deep Learning models. This relatively new framework has been gaining traction so we decided to make sure Valohai supports it well. One of the benefits over competing frameworks has been CNTK’s ground up support for multi-node, multi-GPU training, something that for instance TensorFlow has been struggling to tackle well. If you are doing work on really large datasets, you should maybe give it a try.

Read more [fa icon="arrow-right"]

Synthetic Training Dataset with Unity

[fa icon="calendar'] Nov 13, 2018 1:23:29 PM / by Ruksi posted in Unity, Machine Learning Tools

Synthetic data is artificially created information rather than recorded from real-world events. A simple example would be generating a user profile for John Doe rather than using an actual user profile. This way you can theoretically generate vast amounts of training data for deep learning models and with infinite possibilities.

Read more [fa icon="arrow-right"]

GDPR and its Effects on Machine Learning Based Decisions

[fa icon="calendar'] Oct 31, 2018 8:23:56 AM / by Joanna Purosto posted in GDPR+ML

You might have heard that every individual subject to automated decision making by machine learning models has a right to an explanation of the result. I bet you feel drops of sweat forming on your forehead when you receive an inquiry from a manager saying that he needs details about how a certain decision was made. If thinking about this scenario gives you chills, you are in the right place. Read further and learn how to tackle the transparency issue.

Read more [fa icon="arrow-right"]

What to Store from a Machine Learning Experiment

[fa icon="calendar'] Oct 23, 2018 8:46:02 AM / by Eero Laaksonen posted in Machine Learning Experiment, Machine Learning Infrastructure

When meeting with teams that are working with machine learning today, there is one point above everything else that I try to teach. It is the importance of storing and versioning of machine learning experiments and especially how many things there actually are that need to be stored.

Read more [fa icon="arrow-right"]

Machine Learning Orchestration for Free

[fa icon="calendar'] Oct 16, 2018 8:57:41 AM / by Fredrik Rönnlund posted in Machine Learning Infrastructure, Valohai, Free tier

You know what really grinds my gears? When I have a deep learning model that I want to train and I have to SSH into my AWS instance, install all the drivers and libraries, run my code and then forget to shut down my machine! Once, I ended up forgetting one up over the weekend that cost my employer over $10 000!!!

Read more [fa icon="arrow-right"]

New Release: Managing Your Experiments Just Got Easier

[fa icon="calendar'] Oct 12, 2018 2:01:55 PM / by Ruksi posted in release, docker, training

Recreating experiments inside Valohai could be a whole lot easier and we’ve heard your cries!

With the latest release, live today, whenever you copy an old experiment and want to re-run it Valohai now copies the tags and title over from the previous experiment. Tags are also now automatically propagated down to individual executions when you create a task with several ones e.g. during a hyperparameter sweep.

We also improved the creation of new experiments. When creating a new execution you now have a dropdown for selecting your Docker image. We pre-fill the filterable box with our list of recommended Docker images but you can naturally point to any custom made Docker image as well. And naturally, if you have defined a default one in your valohai.yaml file that will be the default one selected from the get-go.


We have also fixed several bugs and made a handful of smaller fixes in the UI and the API that you can read more about in the patch notes.

Read more [fa icon="arrow-right"]