Cajal course in computational neuroscience 2018
Recently I attended a summer course in the Champalimaud Centre for the Unknown, Lisbon, Portugal. The course was called ‘Cajal course in computational neuroscience 2018’. As the name suggests, it taught the central ideas, methods, and practices of modern computational neuroscience through a combination of lectures and hands-on training.
During the course, I gathered many useful skills that will help guide my PhD project while also preparing me for my future career. Moreover, I had exceptional experience in exploring Lison - its great weather, sea beaches and city life.
The course consisted of lectures on various topics of experimental and computational neuroscience delivered by distinguished international faculty and hands-on training through exercises and project work. It was a 3-week long course with the first week covering various computational methods and the last two weeks covering the use of those methods to do neuroscience research.
The lectures covered the most recent advancements in various fields - sensory processing, cognitive computations, machine learning and so on. I enjoyed the machine learning lectures the most, specially, transfer learning. I learned a lot about the most recent applications of Bayesian statistics in reinforcement learning, for example, Q-learning. I also learned about the models of memory networks, for example, Hopfield network.
The lecture materials included both familiar and unfamiliar topics. For example, I knew most of the literature that was covered during the lectures on encoding models. However, I knew very little about the advanced dimensionality reduction methods, like factor analysis. But after attending the lectures, I felt I had a firm grasp of those seemingly unfamiliar methods.
Research projects were proposed by faculty members in the first week of the course. The choice of projects was very diverse ranging from analysis of neural data to theoretical work on computational principles underlying brain function. I worked on a research project in a team of three to analyze neural data from medial entorhinal cortex.
Medial entorhinal cortex (mEC) has been studied extensively for its role in memory and spatial navigation. This lead to the discovery of various classes of cells in mEC that encode for a particular metric (such as, location, head direction or speed of an animal) of spatial information. For example, grid cells encode place by firing periodically in the place where the animal is located. Head direction cells fire when the animal’s head is oriented towards certain direction. Similarly, border cells and speed cells encode for the boundary of the place or the speed of the animal and so on. While cells in mEC would be classified based on one spatial metric, recent studies suggest that individual cells in mEC might code for more than one of the spatial metrices.
Hardcastle and colleagues have shown that the activity of only a small fraction of cells in mEC can be explained by a single spatial metric. For explaining the activity of most cells, consideration of multiple metrices is necessary, for example, position, speed, head direction or, matric related to the internal environment of the animal, like theta rhythm.
In our project work we decided to examine the mixed selectivity of cells in mEC while the animal was doing a novel two-frame avoidance task, contrary to the classic open field foraging task. To perform the two-frame place avoidance task, the animal has to remember the place to be avoided to keep away from the shock. In the two-frame avoidance task, there are two shock zones; one of them remains static relative to the arena, and the other remains static relative to the room while moving with the arena. So, the animal has to produce representation of two spatial frames and remember specific places in them in order to be able to perform the task. The animals were really good in the task and find food pellets scattered in the arena successfully. Learning is very fast - animals learn the task within the fifth session. We investigated the grid, head direction and speed selectivity of cells in three task conditions: static arena, rotating arena and again static arena. After analysing the data, we found some exciting preliminary results which will be followed on by Christina Savin lab.
To have a clear picture of life during the course, now I want to give an account of my daily schedule during the course. Every weekday would begin by having breakfast in the hotel at eight o’ clock in the morning, starting off for the lectures at half past eight and being present for the lectures by nine o’ clock. We would usually attend two lecture sessions in the morning. After lunch there would be exercises and project work. Each working day would end with dinner at eight o’ clock in the evening. Then we would go back to the hotel again. Because all the attendees were staying in the same hotel, we would often gather in the hotel lobby and have fruitful discussion after work. On some weekdays, we would go out in the city to have drinks. For the weekends there were always some pre-planned activities like boat tour, city tour, beach visit, surfing and so on.
On a few occasions, we were also invited to have dinners with the speakers in the city. Those days were very special to me, as this was an opportunity to learn about the speakers personally. I would often ask them about their PhD life and how they progressed later in their career. The scientists that I went to have dinner with were: Professor Mate Lengyl (Bayesian statistics), Professor Maneesh Sahani (auditory neuroscience – computation), Professor Jennifer Linden (auditory neuroscience – experiment), Professor Eero Simoncelli (computational Vision), Professor Tony Movshon (vision) and Professor Matthias Bethge (deep learning).
I must not forget to mention about my one-to-one conversation with Professor Eero Simoncelli. He is a world expert in visual and auditory neuroscience. We had a very engaging discussion for about two hours. I told him about my research and asked him many questions about it. He had some very important inputs about my PhD project.
My undergraduate degree was on genetics and biotechnology which taught more biology than computational courses. I am self-taught on mathematical modelling and machine learning. It was only during the MSc in Neuroscience at Oxford that I could finally have some formal education in computational neuroscience. However, I still felt the lack of knowledge on quantitative and theoretical neuroscience - Cajal course definitely helped in filling in some of these gaps. The CAJAL course was very relevant to my current research on modelling spiking activity of neurons in auditory midbrain and auditory cortex.
When I think about last summer, I think about Lisbon. Not only that I have learned a lot, I have made many friends and met many colleagues. On one hand, I have become familiar with a lot of topics of computational neuroscience and on the other hand, I have met a lot of potential collaborators who are probably going to lead the field in the upcoming future.
Acknowledgement:
The Boehringer Ingelheim Fonds travel grant financed my expenses for the course.
During the course, I gathered many useful skills that will help guide my PhD project while also preparing me for my future career. Moreover, I had exceptional experience in exploring Lison - its great weather, sea beaches and city life.
The course consisted of lectures on various topics of experimental and computational neuroscience delivered by distinguished international faculty and hands-on training through exercises and project work. It was a 3-week long course with the first week covering various computational methods and the last two weeks covering the use of those methods to do neuroscience research.
The lectures covered the most recent advancements in various fields - sensory processing, cognitive computations, machine learning and so on. I enjoyed the machine learning lectures the most, specially, transfer learning. I learned a lot about the most recent applications of Bayesian statistics in reinforcement learning, for example, Q-learning. I also learned about the models of memory networks, for example, Hopfield network.
The lecture materials included both familiar and unfamiliar topics. For example, I knew most of the literature that was covered during the lectures on encoding models. However, I knew very little about the advanced dimensionality reduction methods, like factor analysis. But after attending the lectures, I felt I had a firm grasp of those seemingly unfamiliar methods.
Research projects were proposed by faculty members in the first week of the course. The choice of projects was very diverse ranging from analysis of neural data to theoretical work on computational principles underlying brain function. I worked on a research project in a team of three to analyze neural data from medial entorhinal cortex.
Medial entorhinal cortex (mEC) has been studied extensively for its role in memory and spatial navigation. This lead to the discovery of various classes of cells in mEC that encode for a particular metric (such as, location, head direction or speed of an animal) of spatial information. For example, grid cells encode place by firing periodically in the place where the animal is located. Head direction cells fire when the animal’s head is oriented towards certain direction. Similarly, border cells and speed cells encode for the boundary of the place or the speed of the animal and so on. While cells in mEC would be classified based on one spatial metric, recent studies suggest that individual cells in mEC might code for more than one of the spatial metrices.
Hardcastle and colleagues have shown that the activity of only a small fraction of cells in mEC can be explained by a single spatial metric. For explaining the activity of most cells, consideration of multiple metrices is necessary, for example, position, speed, head direction or, matric related to the internal environment of the animal, like theta rhythm.
In our project work we decided to examine the mixed selectivity of cells in mEC while the animal was doing a novel two-frame avoidance task, contrary to the classic open field foraging task. To perform the two-frame place avoidance task, the animal has to remember the place to be avoided to keep away from the shock. In the two-frame avoidance task, there are two shock zones; one of them remains static relative to the arena, and the other remains static relative to the room while moving with the arena. So, the animal has to produce representation of two spatial frames and remember specific places in them in order to be able to perform the task. The animals were really good in the task and find food pellets scattered in the arena successfully. Learning is very fast - animals learn the task within the fifth session. We investigated the grid, head direction and speed selectivity of cells in three task conditions: static arena, rotating arena and again static arena. After analysing the data, we found some exciting preliminary results which will be followed on by Christina Savin lab.
To have a clear picture of life during the course, now I want to give an account of my daily schedule during the course. Every weekday would begin by having breakfast in the hotel at eight o’ clock in the morning, starting off for the lectures at half past eight and being present for the lectures by nine o’ clock. We would usually attend two lecture sessions in the morning. After lunch there would be exercises and project work. Each working day would end with dinner at eight o’ clock in the evening. Then we would go back to the hotel again. Because all the attendees were staying in the same hotel, we would often gather in the hotel lobby and have fruitful discussion after work. On some weekdays, we would go out in the city to have drinks. For the weekends there were always some pre-planned activities like boat tour, city tour, beach visit, surfing and so on.
On a few occasions, we were also invited to have dinners with the speakers in the city. Those days were very special to me, as this was an opportunity to learn about the speakers personally. I would often ask them about their PhD life and how they progressed later in their career. The scientists that I went to have dinner with were: Professor Mate Lengyl (Bayesian statistics), Professor Maneesh Sahani (auditory neuroscience – computation), Professor Jennifer Linden (auditory neuroscience – experiment), Professor Eero Simoncelli (computational Vision), Professor Tony Movshon (vision) and Professor Matthias Bethge (deep learning).
I must not forget to mention about my one-to-one conversation with Professor Eero Simoncelli. He is a world expert in visual and auditory neuroscience. We had a very engaging discussion for about two hours. I told him about my research and asked him many questions about it. He had some very important inputs about my PhD project.
My undergraduate degree was on genetics and biotechnology which taught more biology than computational courses. I am self-taught on mathematical modelling and machine learning. It was only during the MSc in Neuroscience at Oxford that I could finally have some formal education in computational neuroscience. However, I still felt the lack of knowledge on quantitative and theoretical neuroscience - Cajal course definitely helped in filling in some of these gaps. The CAJAL course was very relevant to my current research on modelling spiking activity of neurons in auditory midbrain and auditory cortex.
When I think about last summer, I think about Lisbon. Not only that I have learned a lot, I have made many friends and met many colleagues. On one hand, I have become familiar with a lot of topics of computational neuroscience and on the other hand, I have met a lot of potential collaborators who are probably going to lead the field in the upcoming future.
Acknowledgement:
The Boehringer Ingelheim Fonds travel grant financed my expenses for the course.
Comments
Post a Comment