make project 1 training focused

README.md

@@ -9,26 +9,26 @@ Code Challenges submissions should demonstrate creativity, understanding of the
 
 # List of Ideas
 The following ideas were created with feedback from contributors and collaborators. Project ideas are not listed in any particular order. Submissions need not come from the below list, but should have reasonable relevance to the caMicroscope organization and its goals. Please feel free to discuss these or other project ideas on the email list or Slack group.  
-** More ideas are upcoming **
 
 ***
 
-## **[1] Machine Learning Assistant
+## [1] Machine Learning Training Assistant
 
-Many user actions in pathology data collection are mostly independent. Annotation, classification, and user review are produced largely by an expert opinion without use of machine learning tools. To augment this workflow, caMicroscope has added some specific tools which can assist with or validate annotations and classifications. However, at this time the tools are separate enough to often be practically slower than not using them.
-This project, the Machine Learning Assistant, is to consider the needs of the user as well as the complexity to automatically and unobtrusively assist data collection by pathologists and other researchers. 
+Many user actions in pathology data collection are mostly independent. Annotation, classification, and user review are produced largely by an expert opinion without use of machine learning tools. To augment this workflow, caMicroscope has added some specific tools which can assist with or validate annotations and classifications. However, at this time use of the tools usually require a pretrained model in order to use. There is a workflow for creating labeled dataset images and training in caMicroscope, but it is difficult to use in practice.
+This project, the machine learning training assistant, would be both to consolidate and improve the user experience of the training workflow, as well as to find ways to improve performance and time taken for predictions.
+This project would likely involve adding a microservice to allow training to run on a server or third party platform based upon configuration.
 A significant amount of this project would be user experience focused, specifically finding ways to quickly provide insight using existing machine learning tools. To augment this, the project may include additional implementations, as well as runtime analysis to determine what information can be considered at a given time.
 
 Difficulty: Hard
 Project Length: Long
 
 Primary Project Contact: Ryan Birmingham
 
-Source Code: https://github.com/camicroscope/camicroscope 
+Source Code: https://github.com/camicroscope/camicroscope and https://github.com/camicroscope/caracal
 
 Code Challenge: Create a frontend web application which uses tensorflow-js to provide some form of analysis on a user-supplied image unobtrusively. This should function on multiple timescales as possible, so that some information can be displayed immediately, while other slower-running calculations can be returned on completion.
 
-## **[2] Dicom Support**
+## [2] Dicom Support
 
 caMicroscope supports openslide compatible file types for its tiling engine. While many other formats are often requested to work without conversion, one of the most common suggestions that research groups and governmental organizations ask for is DICOM Whole Slide Imaging (See https://dicom.nema.org/dicom/dicomwsi/). This format follows the DICOM standards as are common in radiology, so this would help promote interdisciplinary research. This project would require creating an alternate tiling engine microservice which can efficiently serve regions of a slide from DICOM WSI format into either a simulated deepzoom image or the IIIF url pattern format (see https://iiif.io/).
 
@@ -41,9 +41,9 @@ Primary Project Contact: Tony Pan
 
 Code Challenge: Create a microservice in a language/framework of your choice which can serve metadata about any DICOM file.
 
-## **[3] Multi-channel Imaging Support
+## [3] Multi-channel Imaging Support
 
-## **[4] Collection and Study Management 
+## [4] Collection and Study Management 
 
 Mentors: Nan Li 
 
@@ -59,7 +59,7 @@ Source Code: https://github.com/camicroscope/caMicroscope and https://github.com
 
 Code Challenge: Use object-oriented design to create a simple webapp which provides the operation to manipulate the collection in MongoDB. The code challenge should focus on developing the backend in REST APIs which uses Node.js to operate mongoDB and a simple frontend to proof the contributors understanding the basic architecture of web application. 
 
-## **[5] Modularization of Components 
+## [5] Modularization of Components 
 
 Mentors: Nan Li 
 
@@ -75,7 +75,7 @@ Source Code: https://github.com/camicroscope/caMicroscope
 
 Code Challenge: A simple webapp demonstrating Web Components 
 
-## **[6] Eaglescope Automatic Configuration**
+## [6] Eaglescope Automatic Configuration
 
 Eaglescope is a web application for exploratory analysis on high dimensional datasets, especially biomedical datasets. This tool has been designed primarily to focus on cohort identification, that is to identify a set of criteria which produce distinct or interesting data. Right now, in order to create a dashboard, the entire layout needs to be specified in a configuration manifest. However, this has the side effect of requiring that users already understand the data well enough to select which possible visualizations would be most interesting.
 This project, Eaglescope Automatic Configuration, aims to provide instant statistical insight into which fields or combinations of fields can be best represented in the various different visualizations implemented in eaglescope. This would have the added bonus of letting a user quickly explore a new dataset without writing any configuration. This has been proposed as a short project, and would focus on classical statistical methods. 
@@ -90,17 +90,19 @@ Source Code: https://github.com/sharmalab/eaglescope
 Code Challenge: Create a web application which takes in a user-supplied file (e.g. csv) and uses SVD and/or dimensionality reduction to determine which feature(s) are most explanatory.
 
 
-## [7] Pathology Game
+## [7] User Driven Pathology Validation
 
 Primary Mentor: Ryan Birmingham
 
-Overview: caMicroscope has a viewer component and support for machine learning models. Thus, we have the components to build an alternate version of the viewer to host a pathologist vs machine learning model game. If the model is trusted, this application should function as a way for a human to test their observation skills in pathology. Alternatively, if a model is unvalidated, this would work as a way to test the model.  Practically, this would involve creating at least one interactive way to numerically compare a human input with a machine learning model classification or segmentation.This would serve as a demo of caMicroscope, a way to validate machine learning models, and could be fun or an informal training exercise for pathologists.
+Overview: caMicroscope has a viewer component and support for machine learning models. Thus, we have the components to build an alternate version of the viewer to host a pathologist vs machine learning model visual and numeric annotation comparison environment.
+If the model is trusted, this application should function as a way for a human to test their observation skills in pathology. Alternatively, if a model is unvalidated, this would work as a way to test the model.  Practically, this would involve creating at least one interactive way to numerically compare a human input with a machine learning model classification or segmentation. Strong proposals would demonstrate at least a few such ways of showing this comparison both numerically and visually.
+This would serve as a demo of caMicroscope, a way to validate machine learning models, and could be alternate kind of training exercise for pathologists.
 
-Current Status: New App
+Current Status: New frontend application
 
 Required Skills: UX, JavaScript, TensorFlow
 
-Code Challenge: Make a clone/mvp of a game similar to geoguessr, focusing on the user interaction and scoring.
+Code Challenge: Make a clone/minimum viable product of a game similar to geoguessr, focusing on the user interaction and scoring.
 
 Difficulty: Hard
 Project Length: Long