Within the past month, four Silo.AI employees completed their doctoral studies and received their PhDs. We’re celebrating the big milestone this Friday, and want to recognize their academic accomplishments with this blog post (and some bubbly at the office).
1. Congratulations! How do you feel?
Sami Remes: Mostly relieved, after having spent numerous evenings and weekends working to finalize my thesis, that now this is finally done.
Luiza Sayffullina: I feel after the discussion with the opponent that actually my work is valuable for the society.
Alexander Grigorevskiy: I feel that I am gradually becoming more relaxed. On the other hand I feel that now there is an emptiness in the place of long-term PhD project. I need to think how to fill this emptiness.
Jesus Carabaño: Liberated, but also confused… I cannot use my thesis as an excuse anymore!
2. What has been most rewarding in the PhD studies and process?
SR: Being able to learn a lot, to go deep and understand the methods that I’ve been working with, as well as being able to attend conferences and talk with many experts in their fields
LS: All kinds of opportunities, including going to an exchange period in the top research labs, becoming friends with the people in the international scientific community and becoming truly independent in my research.
AG: I enjoyed many things: studies, doing research, traveling, meeting with very intelligent people. Even difficult moments now look like a good experiences. I think I became more mature in many senses – this is probably the most rewarding aspect.
JC: All the learning and experiences, especially my mental evolution from shallow and gullible to rational and observant. Nowadays I don’t trust anything without a sound mathematical proof.
3. Why did you choose this path to become a PhD and continue your studies?
SR: I had started working as a research assistant in the research group already during my Bachelor’s studies, and I wanted to continue doing more research and becoming better and more independent in the research process.
LS: It was clear to me after my Master’s that I want to continue to study Machine Learning on a deeper level and to do research.
AG: After my master’s degree this was the most straightforward way for me. I knew that I enjoyed doing intellectual work, but at that point I did not feel the big difference between industry and academia. So, in the beginning I had some doubts. After about one year of PhD I was sure that I was on the right track.
JC: I took a wrong turn in some shady corridor at the university, where I found my supervisor. He lured me with shiny brand new GPUs and at some point I was too deep into that mess to escape, so I just went ahead and finished the PhD instead.
4. What does doing research mean to you?
SR: Discovering new problems and proposing new solutions to deal with them
LS: Coming up with new impactful problems which can be solved the best with your expertise area.
AG: Research is about asking important questions and finding answers. But now I think it is also a way my mind works. Not necessary I am contemplating some technical issues, but there is always something in my mind I am curious about and want to find answers to.
JC: Embracing the scientific method as the my true religion.
SAMI REMES – PhD in Machine Learning
Sami is a senior machine learning expert with 8+ years of experience in researching and applying machine learning. During his academic career Sami has published at the NIPS conference and other well-established machine learning venues.
Sami recently defended his dissertation entitled “Modelling non-stationary functions with Gaussian processes” at the Department of Computer Science at the School of Science at Aalto University. The dissertation presents non-stationary kernel functions for Gaussian processes, which can be utilized in modelling changing dependencies between different measurements over time. Gaussian processes (GP) are a central piece of non-parametric Bayesian methods, which allow placing priors over functions in settings such as classification and regression.
The choice of the kernel is crucial when applying Gaussian processes. However, the commonly used standard kernels often offer unsatisfactory performance due to making the assumption of stationarity. Sami’s thesis presents approaches in modelling non-stationarity from two different perspectives in Gaussian processes.
At Silo.AI, Sami works as AI Scientist and has been part of various machine learning and deep learning projects, including modeling expert knowledge for a financial sector client in order to speed up, and in the future, to automate the invoicing process.
Dissertation available online at: https://aaltodoc.aalto.fi/handle/123456789/40156.
LUIZA SAYFULLINA – PhD in Machine Learning for Natural Language Processing
Luiza is a senior machine learning expert with 7+ years of machine learning experience and a deep understanding of Natural Language Processing for English and Finnish language. During her academic career Luiza has first-authored eight papers and has been active in organizing machine learning study groups since 2016.
Luiza just published her dissertation “Machine Learning Methods for Classification of Unstructured Data” at the Department of Computer Science at the School of Science at Aalto University. Her research focuses on analyzing unstructured data, in other words, free-form text. With its two applications, Android malware classification and soft skill mining, Luiza uses NLP to analyze job candidate resumes and Android application files.
At Silo.AI, Luiza works as AI Scientist with a focus on NLP projects ranging from low resource text classification, information extraction and summarization to speech-to-text applications.
Dissertation available online at: https://aaltodoc.aalto.fi/handle/123456789/40155.
ALEXANDER GRIGOREVSKIY – PhD in Machine Learning
Alexander is senior machine learning expert with 10+ years of experience in machine learning research, industrial machine learning projects, consulting and software development. During his academic career, Alexander has authored 10 articles with nearly 150 citations and accumulated deep knowledge in several areas of machine learning and artificial intelligence.
Alexander’s dissertation “Advances in Randomly-Weighted Neural Networks and Temporal Gaussian Processes” was recently published at the Department of Computer Science at the School of Science at Aalto University. In his dissertation, Alexander discovered that back-propagation is not the only way to train neural networks as randomly assigning the weights of the first layers and using the ordinary least-squares often provides much faster but comparable alternative. In addition, the dissertation covers time series prediction and modeling using probabilistic methods.
Alexander’s dissertation shows that randomly weighted neural networks are favorable in situations when fast training time is required. In that sense, they are a nice addition to the collection of machine learning models. They also allow to better understand conventional neural networks.
At Silo.AI Alexander works as AI Scientist, with a proven track record of delivering practical machine learning implementations, including fraud detection for global audit company. His dissertation is relevant for applying AI to business problems, as time series modeling and predictions can support decision making at many companies.
Dissertation available online at: https://aaltodoc.aalto.fi/handle/123456789/40261?locale-attribute=fi.
JESUS CARABAÑO BRAVO – PhD in High Performance Computing
Jesus is a skilled machine learning expert with experience in computer vision and machine learning combined with software development. During his doctoral studies, Jesus has also focused on High Performance Computing, and attended several academic conferences as a presenter, including International Supercomputing Conference.
Jesus recently defended his dissertation “A Compiler Approach to Map Algebra for Raster Spatial Model “ at the Department of Computer Science at the Faculty of Science and Engineering at Åbo Akademi University. In one sentence, his dissertation proposes a compiler approach to map algebra. The idea of the approach is that spatial modellers write sequential scripts in the map algebra formalism and the compiler parses it to generate parallel code that runs efficiently on a modern highly heterogeneous computer architectures. The thesis presents the prototype compiler, which is able to handle large volumes of data very efficiently.
At Silo.AI, Jesus works as AI Scientist with a particular focus on machine vision projects.
Dissertation available online at: https://www.doria.fi/handle/10024/170111.