Replies: 1 comment
-
Paul Prae's Rough NotesGoalsDiscovery in brain imaging and translating that for the community. Focused on brain dynamics, multimodal fusion combining data sources together. Deep learning and blind separation. Brain is complex so we want to discover patterns so we can know how to predict. Working on applying these to mental health like autism and schizo. Brain science with Alzheimers. As data grows we need to manage the data. Neuroinmofratics tools are important here. COINSTAC helps with data protection and privacy. Trying to position itself to complement open data. Interface that with and augment with hidden data like those locked in institutions such as with clinical data sets. In the future interfacing with EHR systems. How do these efforts cascade down to help everyday people? This is all motivated by helping individual lives. GrantsCOINSTAC 2.0 project is in year two. Five year extension. This is coupled with a Vermont group on substance abuse and NIDA. Can interface with substance abuse group in ENIGMA. Jess is building this consortia. Lots of international groups contributing with ENIGMA. Some people held back their data, but COINSTAC gave them the confidence to pull in their data. Goal is to make ENIGMA more effective. Another way we are positioning our data involves Vault. Organizing data is hard. Need to make it easy to plug in with approved data set. Data sharing. ENIGMA is may of 2024. Could work towards that in July of next year. We are even leveraging satellite data and environmental factors. Gunter Shuman is working on mental health. Great international group. RO1 are five year project at five million dollars that we lead. Halfpipe is something like a sub award that we don’t own. Smaller like a few hundred thousand. Another university may own them. Supplements are smaller things like $100k. These can be added on to RO1. Different computations tend to map to goals of particular grants or partnerships. We do have a list of tasks from the grants and partners and funders. In general we can act in spirit so it gives us some creativity. Differentiating ourselvesVBM and Freesurfer are a combination of commonly used, low hanging fruit, with differentiating. These need to be hardened. Group ICA with FMRI. Unit testing is hard. Integration testing may be easier to like make sure they run on COINSTAC. Like a provision where we run a test suite. Testing can force us to have a more standardized interface. Need more working examples that are easy accessible. Automate the running of examples after changes occur. Data is not supposed to be in containers. Docker should ignore data in test containers. Data is often in the repo. Masked data is sometimes in containers. GUI integration is a huge pain. Need better interface between GUI and simulator. Prioritization and Project ManagementNeed focused sprint goals. Ross got dev working recently for Kelly and Ross. Dylan wants help voting on what is most important. Ross would like us to know why thins are a priority on Thursday. Map things to benefits to all stakeholders. Need more dev to management meeting. It may need to be more like those agile standup with the business people involved. Better balance urgency with long term value and with technical debt. Roll things up into Epics or Themes. Have people vote on these Epics and Themes. They are like collections of Issues. Can we vote with weighted voting in Slack by stakeholder. Number of tags can also be a way to rank. Really need to vote on tags themselves too? These overall buckets and collections that are represented as tags are hard to define and manage. A tag could be around “community support”. Things like documentation will help with engaging the broader community. Can tag around industry trends like “data privacy.” Could open source community engagement be a Supplement grant? First need to communicate more effectively with leadership. Then codify it. Scale and ProductivityFOCUS and say no to almost everything to accomplish more over time. Team ADHD of devs is our problem. But leadership ADHD of where tasks often come from is not in our control. We need ownership there from Vince and Sandeep. Vince is big into tapping into the open source community. Doesn’t require more funding in this route. DAO’s and open science could help here. Partnerships could help us scale out to more people too. Still need a core maintenance and support team plus a group internally that is more about long term strategy and vision. Need to unlock outside collaborators. Organizing a COINSTAC computation hackathon. Could use this as a working backwards story. What needs to be done to support a productive, fund event like this? Don’t have time to pursue more research grants. Need to scale grant writing and grant management. Ties to people and project management which is consuming. Where one hat at a time as long as we’re able to. Context switching has a real cost. A better UX in the app itself kind of acts as documentation for computation authors. Like in Javier’s designs how he visualized the overall process. Javier’s presentation got a lot of engagement from the PI’s! Our new feature lists need to be clustered. Vince has many features and issues clustered in his mind. First thing we need is a first set of core computations. How does COINSTAC pre-processing differ from what Brainforge does? By default, anything labeled as “core” is highest priority such as the top most useful or popular computations. It may be best for the pre-processing containers in Brainforge to be aligned with COINSTAC. Make sure to reduce redundancy. Share computations that are the same. Need to have a clear integration and interface between the two. Jeremy would be ahead of all of our neuroinformatics tools. An MLOps engineer could be shared between Brainforge and COINSTAC to help with things like sharing and managing preprocessing computation containers. Grants pay for innovation and new things. Need other funding sources for maintenance and support. Maybe commercialization efforts create that kind of team. Video tutorials would be great. Can we define our “core” features as user stories and use those to drive the documentation we write? How we market the tools is also the features we focus on and what we prioritze and document. Some devs look at other tools as more flexible and easier to use. Need comp author engagement to be fully automated. Contributing to COINSTAC is centralized and needs to be decentralized. We need to aggressively adopt existing open source software that already does things we’re doing. Maintaining and developing software is really hard, anyway we can get others to build features for us we want to. We need to be more clear how we differentiate ourselves. Need to better highlight our core value. The workflow and UI we have is a big differentiating factor. Differential privacy over neuroimages is another part of our niche. We need a go-to-market research kind of strategy document. Open Differential Privacy is a thing at federal levels like on census data. Current approaches don’t provide meaningful approaches. REDCAPREDCAP is a database intended for clinical studies. REDCAP is eager at getting integrated with REDCAP. Like GSU, they purchase data from qualtrics. REDCAP is a for profit business. They have private IP. There is open source software integrating with it though. Redcap use case: an API for pipelines. Vaults and pipelines go together. Can we standardize pipelines so we can pull data out of them. If we put data into pipeline X, we know how to format it for things like large scale extractions. PeopleSunitha working on COINSTAC computation. Neuroscience researcher. Anand is from Rutgers and is here to work on Differential Privacy. Algorithm design. Sumanth is in CV, healthcare, MS in big data Questions / Parking LotDoes it makes sense that I be the engineering lead across the org? Per Jeremy’s goals of a centralized team. Ask team where we need the most help in general, where they’ve found me most helpful in the past, where they would see most helpful in the future, and share my answers to these along with what I enjoy the most.
We are open to engaging with clinical trials. Build tools to open up more options here. Need to make it more clear how team composition and processes enables everything. It can unlock many ways we are blocked, like new features vs support vs new build/re-architecture. Full-time person for MLOps. Partner with FourthBrain. Data (maps to trained models), code (COINSTAC - repo version and packaged version - and computations/algorithms), model version control. Testing. Managing input spects. Managing trained models. This helps with repeatability. Need to keep the knowledge alive and active like to keep computations working and useful. Need standardization and less variability. I need to get more of my previous work to a good stopping point like documentation. I need to find a way to minimally keep technical project management moving along. Be the connector and guide. Why more money? Produce more research more easily. Maintained devs because we don’t have to where so many hats, less stress, less context switching. Can PI’s vote on issues with parrots? Can we map tasks to how many people or grants or organizations they help? Need to get slides from Jess and Sandeep |
Beta Was this translation helpful? Give feedback.
-
Place your meetings notes as comments here.
Beta Was this translation helpful? Give feedback.
All reactions