Practicum Pride: Recology
Paul Kim came to the MSDS program with a Bachelors of Arts in Radio and Television from Northwestern University and Katja Wittfoth came to the MSDS program after working for 10 years at Bayer. Continue reading to learn more about their practicum experience at Recology!
Can you tell us a bit about Recology?
PAUL — Recology is a resource recovery company headquartered here in San Francisco. If you live in the city, you most likely have seen the logo on the side of your waste, recycle, and compost bins. Recology is working toward a no-waste solution by 2020, which means no material sent to landfills. The company is 100% employee-owned and is the largest employee-owned company in the industry. They also have an artist-in-residence program where artists will come and make pieces out of recovered disposed items.
Can you describe the project(s) that you are working on?
PAUL — We are working on a computer vision project, trying to identify contaminants in different bins. For example, if there is non-compostable material in the compost bin, our model should be able to detect it. The goal is to automate the waste auditing process, working towards the 2020 goal of no-waste.
KATJA — Recology is aiming to streamline and automate the identification of contaminants using deep learning algorithm. We are working with a convolutional neural network model which showed 94% accuracy. Currently, we are refining the metrics and turning the model for the set of pictures taken by the truck cameras. In the Spring, we will focus on building the application to connect the high-definition truck cameras with our model.
How are you applying the knowledge gained from the program to your practicum? Is there a particular class that has been the most helpful?
PAUL — Of course, Machine Learning has been the most helpful course so far. Though we are not using algorithms taught in the ML course, the general process of tuning parameters, finding learning rates, etc. has been crucial to our work.
KATJA — The skills we gained in Statistics, Python, and Distributed Computing were extremely helpful to navigate through the current model which was built by the last cohort’s interns. We also applied knowledge from our Machine Learning class to tune the algorithm while playing with different hyperparameters and validation metrics. During the next module, we will enroll in Distributed Data Systems and Advanced Machine Learning, which will help us make the model ready for production.
What is it like working with professional data scientists?
KATJA — There are actually no data scientists on the team, which has its advantages and disadvantages. While it would be nice to learn from experienced data scientists, it is really rewarding to be the driving force of the data innovation.
PAUL — We are the main data science resource at Recology. However, our manager is highly interested in data science and is looking to expand data science at the company. Recology has a lot of data, as well as the desire to leverage that data. It’s been very interesting to experience firsthand what it’s like to work with non-technical people on a very technical project; not only have we been using our coding skills, but we’ve also been honing our communication skills.
What is the biggest challenge you’ve faced in your practicum?
KATJA — The biggest challenge we faced is working with someone else’s code. The original model was written by the students from last year’s cohort. We took some time to understand their model before we started to further tune the algorithm. It was very helpful to be able to reach out to the alumni and receive not only answers to our questions but also recommendations for the model experimentations.
PAUL — I agree with Katja, that was challenging! Another challenge for us was applying what we learned in class to real-world situations. For example, we all learned how to launch an AWS instance, but we didn’t know how to tunnel into the machine to launch a Jupyter Notebook instance.
Are there any cool perks?
KATJA — One of the perks is the possibility to work remotely. We usually work half the time from Recology and another half from campus. On campus, we have access to our faculty as well as to other students who work on the deep learning projects remotely.
For more information on the MS in Data Science program:
