So I started this blog when I was working as a data scientist in an industry job. It so happened that 6 months ago I got hired as a consultant at the Howard Hughes Medical Institute to work on machine learning based analysis of large-scale neural recordings at the Pachitariu lab https://www.janelia.org/lab/pachitariu-lab How did that happen? Well, I got really lucky. I worked on calcium imaging data from the Allen Brain Observatory for my MSc thesis. To be honest my MSc thesis was unsatisfactory. I fit a Latent Dirichlet Allocation model to neural data, but couldn't get decent performance on the decoding the stimuli and in general I feel I didn't really understand what I was doing. Then the summer after I graduated I went to a Berkeley summer school on Neural Data Mining. When I applied I wrote that I want to be more rigorous in my research. When I was there I talked to professor Maneesh Sahani about my mishaps and he recommended I take a look at this dataset https://figshare.com/articles/Recordings_of_ten_thousand_neurons_in_visual_cortex_in_response_to_2_800_natural_images/6845348 It's a public dataset with calcium imaging recordings from 10000 neurons in V1. I started trying some methods on the data and wrote to Carsen Stringer, who is an investigator at Janelia. At some point she talked to Marius Pachitariu, who is the head of their lab and they wanted to interview me for a consultant position. And it so happens that I got hired.
Marius liked the idea of Latent Dirichlet kind of clustering of neurons, because neurons can belong to multiple clusters or cell ensembles in that models. But within 3 days on the job, Marius wrote down an algorithm he called Ensemble Pursuit for finding sparse clusters of cells in deconvolved calcium imaging data. For the past 6 months I've been working on implementing this algorithm and using it for neural data analysis. Check out the github of the project https://github.com/mariakesa/EnsemblePursuit It tracks the full development history.
During the course of the past 6 months we discovered that cell ensembles in V1 or more concretely average time courses of neurons grouped together in an ensemble have Gabor receptive fields.
A month ago I took an online course on Reproducible Research from Harvard https://courses.edx.org/certificates/user/839946/course/course-v1:HarvardX+PH527x+1T2019 It changed the way I think about the process of research and I decided to repurpose this blog to be a kind of a lab notebook of my research and ideas. It is good to document your research and thoughts as you progress in a project.
I work with highly accomplished scientists and most of the time I feel intimidated:-D But I decided to keep this blog as an informal documentation of my research and progress. I still don't have a PhD degree (I'm applying in a few months) so I'm still learning. Writing down your research and thoughts as is is honest. All the better if it will inform and inspire somebody. Why not share what you're learning as you're doing your research?
Science is messy. It's complex and demanding. You're learning while you're doing. You're not born as an excellent scientist in the same way you're not born a concert violinist. It takes practice and work. I want to be an excellent neuroscientist and this blog will document my journey.
Marius liked the idea of Latent Dirichlet kind of clustering of neurons, because neurons can belong to multiple clusters or cell ensembles in that models. But within 3 days on the job, Marius wrote down an algorithm he called Ensemble Pursuit for finding sparse clusters of cells in deconvolved calcium imaging data. For the past 6 months I've been working on implementing this algorithm and using it for neural data analysis. Check out the github of the project https://github.com/mariakesa/EnsemblePursuit It tracks the full development history.
During the course of the past 6 months we discovered that cell ensembles in V1 or more concretely average time courses of neurons grouped together in an ensemble have Gabor receptive fields.
A month ago I took an online course on Reproducible Research from Harvard https://courses.edx.org/certificates/user/839946/course/course-v1:HarvardX+PH527x+1T2019 It changed the way I think about the process of research and I decided to repurpose this blog to be a kind of a lab notebook of my research and ideas. It is good to document your research and thoughts as you progress in a project.
I work with highly accomplished scientists and most of the time I feel intimidated:-D But I decided to keep this blog as an informal documentation of my research and progress. I still don't have a PhD degree (I'm applying in a few months) so I'm still learning. Writing down your research and thoughts as is is honest. All the better if it will inform and inspire somebody. Why not share what you're learning as you're doing your research?
Science is messy. It's complex and demanding. You're learning while you're doing. You're not born as an excellent scientist in the same way you're not born a concert violinist. It takes practice and work. I want to be an excellent neuroscientist and this blog will document my journey.
Kommentaarid
Postita kommentaar