Tools for managing your PhD projects, reading lists, notes, and inspiration.

As a researcher, you have to know everything. But knowledge is useless if it cannot be accessed quickly. An easy-to-use method of archiving information makes taking notes effortless and enjoyable.
As a PhD student in Artificial Intelligence, I use Obsidian (https://obsidian.md) to manage my knowledge.

The article has three parts:

1. What is a note, how to organize notes, tags, folders, and links? This section is tool-agnostic, so you can use most of these ideas with any note-taking app.
2. Instructions for using Obsidian, managing notes, reading lists, and useful plugins. This section demonstrates how I use Obsidian, my preferred knowledge management tool.
3. Workflows: How to use Zotero to take notes from papers, manage multiple projects' notes, create MOCs with Dataview, and more. This section explains how to use Obsidian to solve common scientific problems and manage/maintain your knowledge effectively.

This list is not perfect or complete, but it is my current solution to problems I've encountered during my PhD. Please leave additional comments or contact me if you have any feedback. I'll try to update this article.
Throughout the article, I'll refer to your digital library as your "Obsidian Vault" or "Zettelkasten".
Other useful resources are listed at the end of the article.

1. Philosophy: Taking and organizing notes

Before diving into Obsidian, let's establish a Personal Knowledge Management System and a Zettelkasten. You can skip to Section 2 if you already know these terms.
Niklas Luhmann, a prolific sociologist who wrote 400 papers and 70 books, inspired this section and much of Zettelkasten. Zettelkasten means “slip box” (or library in this article). His Zettlekasten had around 90000 physical notes, which can be found here.
There are now many tools available to help with this process. Obsidian's website has a good introduction section: https://publish.obsidian.md/hub/

Notes

We'll start with "What is a note?" Although it may seem trivial, the answer depends on the topic or your note-taking style. The idea is that a note is as “atomic” (i.e. You should read the note and get the idea right away.
The resolution of your notes depends on their detail. Deep Learning, for example, could be a general description of Neural Networks, with a few notes on the various architectures (eg. Recurrent Neural Networks, Convolutional Neural Networks etc..).
Limiting length and detail is a good rule of thumb. If you need more detail in a specific section of this note, break it up into smaller notes. Deep Learning now has three notes:

• Deep Learning
• Recurrent Neural Networks
  - Convolutional Neural Networks

Repeat this step as needed until you achieve the desired granularity. You might want to put these notes in a “Neural Networks” folder because they are all about the same thing. But there's a better way:

The main issue with folders is that they are not flexible and assume that all notes in the folder belong to a single category. This makes it difficult to make connections between topics.
Deep Learning has been used to predict protein structure (AlphaFold) and classify images (ImageNet). Imagine a folder structure like this:

- /Proteins/ 
     - Protein Folding
- /Deep Learning/
     - /Proteins/ 

Your notes about Protein Folding and Convolutional Neural Networks will be separate, and you won't be able to find them in the same folder.
This can be solved in several ways. The most common one is to use tags rather than folders. A note can be grouped with multiple topics this way. Obsidian tags can also be nested (have subtags).

You can also link two notes together. You can build your “Knowledge Graph” in Obsidian and other note-taking apps like Obsidian.

My Knowledge Graph. Green: Biology, Red: Machine Learning, Yellow: Autoencoders, Blue: Graphs, Brown: Tags.

My Knowledge Graph and the note “Backrpropagation” and its links.

Backpropagation note and all its links

Why use Folders?

Folders help organize your vault as it grows. The main suggestion is to have few folders that "weakly" collect groups of notes or better yet, notes from different sources.
Among my Zettelkasten folders are:

My Zettelkasten's 5 folders

They usually gather data from various sources:

MOC: Map of Contents for the Zettelkasten.
Projects: Contains one note for each side-project of my PhD where I log my progress and ideas. Notes are linked to these.
Bio and ML: These two are the main content of my Zettelkasten and could theoretically be combined.
Papers: All my scientific paper notes go here. A bibliography links the notes. Zotero .bib file
Books: I make a note for each book I read, which I then split into multiple notes.

Keeping images separate from other files can help keep your main folders clean.

I will elaborate on these in the Workflow Section.

My general recommendation is to use tags and links instead of folders.

Maps of Content (MOC) 

Making Tables of Contents is a good solution (MOCs).
These are notes that "signposts" your Zettelkasten library, directing you to the right type of notes. It can link to other notes based on common tags. This is usually done with a title, then your notes related to that title. As an example:

An example of a Machine Learning MOC generated with Dataview.

As shown above, my Machine Learning MOC begins with the basics. Then it's on to Variational Auto-Encoders. Not only does this save time, but it also saves scrolling through the tag search section.
So I keep MOCs at the top of my library so I can quickly find information and see my library. These MOCs are generated automatically using an Obsidian Plugin called Dataview (https://github.com/blacksmithgu/obsidian-dataview).
Ideally, MOCs could be expanded to include more information about the notes, their status, and what's left to do. In the absence of this, Dataview does a fantastic job at creating a good structure for your notes.
In the absence of this, Dataview does a fantastic job at creating a good structure for your notes.

2. Tools: Knowing Obsidian

Obsidian is my preferred tool because it is free, all notes are stored in Markdown format, and each panel can be dragged and dropped. You can get it here: https://obsidian.md/

Obsidian interface. 

Obsidian is highly customizable, so here is my preferred interface:

The theme is customized from https://github.com/colineckert/obsidian-things

Alternatively, each panel can be collapsed, moved, or removed as desired. To open a panel later, click on the vertical "..." (bottom left of the note panel).

My interface is organized as follows:

How my Obsidian Interface is organized.

Folders/Search:
This is where I keep all relevant folders. I usually use the MOC note to navigate, but sometimes I use the search button to find a note.

Tags:
I use nested tags and look into each one to find specific notes to link.

cMenu:
Easy-to-use menu plugin cMenu (https://github.com/chetachiezikeuzor/cMenu-Plugin)

Global Graph:
The global graph shows all your notes (linked and unlinked). Linked notes will appear closer together. Zoom in to read each note's title. It's a bit overwhelming at first, but as your library grows, you get used to the positions and start thinking of new connections between notes.

Local Graph:
Your current note will be shown in relation to other linked notes in your library. When needed, you can quickly jump to another link and back to the current note.

Links:
Finally, an outline panel and the plugin Obsidian Power Search (https://github.com/aviral-batra/obsidian-power-search) allow me to search my vault by highlighting text.

Start using the tool and worry about panel positioning later. I encourage you to find the best use-case for your library.

Plugins

An additional benefit of using Obsidian is the large plugin library. I use several (Calendar, Citations, Dataview, Templater, Admonition):
Obsidian Calendar Plugin: https://github.com/liamcain
It organizes your notes on a calendar. This is ideal for meeting notes or keeping a journal.

Calendar addon from hans/obsidian-citation-plugin
Obsidian Citation Plugin: https://github.com/hans/
Allows you to cite papers from a.bib file. You can also customize your notes (eg. Title, Authors, Abstract etc..)

Plugin citation from hans/obsidian-citation-plugin
Obsidian Dataview: https://github.com/blacksmithgu/
A powerful plugin that allows you to query your library as a database and generate content automatically. See the MOC section for an example.
Allows you to create notes with specific templates like dates, tags, and headings.

Templater. Obsidian Admonition: https://github.com/valentine195/obsidian-admonition
Blocks allow you to organize your notes.

Plugin warning. Obsidian Admonition (valentine195)
There are many more, but this list should get you started.

3. Workflows: Cool stuff

Here are a few of my workflows for using obsidian for scientific research. This is a list of resources I've found useful for my use-cases. I'll outline and describe them briefly so you can skim them quickly.
3.1 Using Templates to Structure Notes
3.2 Free Note Syncing (Laptop, Phone, Tablet)
3.3 Zotero/Mendeley/JabRef -> Obsidian — Managing Reading Lists
3.4 Projects and Lab Books
3.5 Private Encrypted Diary

3.1 Using Templates to Structure Notes

Plugins: Templater and Dataview (optional).
To take effective notes, you must first make adding new notes as easy as possible. Templates can save you time and give your notes a consistent structure. As an example:

An example of a note using a template.

### [[YOUR MOC]]
# Note Title of your note
**Tags**:: 
**Links**::

The top line links to your knowledge base's Map of Content (MOC) (see previous sections). After the title, I add tags (and a link between the note and the tag) and links to related notes.
To quickly identify all notes that need to be expanded, I add the tag “#todo”. In the “TODO:” section, I list the tasks within the note.
The rest are notes on the topic.
Templater can help you create these templates. For new books, I use the following template:

### [[Books MOC]]
# Title
**Author**:: 
**Date::
**Tags:: 
**Links::

A book template example.

Using a simple query, I can hook Dataview to it.

dataview  
table author as Author, date as “Date Finished”, tags as “Tags”, grade as “Grade”  
from “4. Books”  
SORT grade DESCENDING

using Dataview to query templates.

3.2 Free Note Syncing (Laptop, Phone, Tablet)

No plugins used.

One of my favorite features of Obsidian is the library's self-contained and portable format. Your folder contains everything (plugins included).

Ordinary folders and documents are available as well. There is also a “.obsidian” folder. This contains all your plugins and settings, so you can use it on other devices.
So you can use Google Drive, iCloud, or Dropbox for free as long as you sync your folder (note: your folder should be in your Cloud Folder).

For my iOS and macOS work, I prefer iCloud. You can also use the paid service Obsidian Sync.
3.3 Obsidian — Managing Reading Lists and Notes in Zotero/Mendeley/JabRef
Plugins: Quotes (required).

3.3 Zotero/Mendeley/JabRef -> Obsidian — Taking Notes and Managing Reading Lists of Scientific Papers

My preferred reference manager is Zotero, but this workflow should work with any reference manager that produces a .bib file. This file is exported to my cloud folder so I can access it from any platform.

My Zotero library is tagged as follows:

My reference manager's tags

For readings, I usually search for the tags “!!!” and “To-Read” and select a paper. Annotate the paper next (either on PDF using GoodNotes or on physical paper).
Then I make a paper page using a template in the Citations plugin settings:

An example of my citations template.

Create a new note, open the command list with CMD/CTRL + P, and find the Citations “Insert literature note content in the current pane” to see this lovely view.

Citation generated by the article https://doi.org/10.1101/2022.01.24.22269144

You can then convert your notes to digital. I found that transcribing helped me retain information better.

3.4 Projects and Lab Books

Plugins: Tweaker (required).
PhD students offering advice on thesis writing are common (read as regret). I started asking them what they would have done differently or earlier.

Time management requires planning. Keeping track of multiple projects and lab books is difficult during a PhD. How I deal with it:

• One folder for all my projects
• One file for each project
  I use a template to create each project

### [[Projects MOC]]  
# <% tp.file.title %>  
**Tags**::  
**Links**::  
**URL**::  
**Project Description**::## Notes:  
### <% tp.file.last_modified_date(“dddd Do MMMM YYYY”) %>  
#### Done:  
#### TODO:  
#### Notes

You can insert a template into a new note with CMD + P and looking for the Templater option.

I then keep adding new days with another template:

### <% tp.file.last_modified_date("dddd Do MMMM YYYY") %>  
#### Done:  
#### TODO:  
#### Notes:

This way you can keep adding days to your project and update with reasonings and things you still have to do and have done. An example below:

Example of project note with timestamped notes.

3.5 Private Encrypted Diary

This is one of my favorite Obsidian uses.
Mini Diary's interface has long frustrated me. After the author archived the project, I looked for a replacement. I had two demands:

1. It had to be private, and nobody had to be able to read the entries.
2. Cloud syncing was required for editing on multiple devices.

Then I learned about encrypting the Obsidian folder. Then decrypt and open the folder with Obsidian. Sync the folder as usual.
Use CryptoMator (https://cryptomator.org/). Create an encrypted folder in Cryptomator for your Obsidian vault, set a password, and let it do the rest.
If you need a step-by-step video guide, here it is:

Conclusion

So, I hope this was helpful!
In the first section of the article, we discussed notes and note-taking techniques. We discussed when to use tags and links over folders and when to break up larger notes.
Then we learned about Obsidian, its interface, and some useful plugins like Citations for citing papers and Templater for creating note templates.
Finally, we discussed workflows and how to use Zotero to take notes from scientific papers, as well as managing Lab Books and Private Encrypted Diaries.
Thanks for reading and commenting :)

Read original post here

As you know, I like to turn my intuition into decision rules (principles) that can be back-tested and automated to create a portfolio of alpha bets. I use one for bubbles. Having seen many bubbles in my 50+ years of investing, I described what makes a bubble and how to identify them in markets—not just stocks.

A bubble market has a high degree of the following:

1. High prices compared to traditional values (e.g., by taking the present value of their cash flows for the duration of the asset and comparing it with their interest rates).
2. Conditons incompatible with long-term growth (e.g., extrapolating past revenue and earnings growth rates late in the cycle).
3. Many new and inexperienced buyers were drawn in by the perceived hot market.
4. Broad bullish sentiment.
5. Debt financing a large portion of purchases.
6. Lots of forward and speculative purchases to profit from price rises (e.g., inventories that are more than needed, contracted forward purchases, etc.).

I use these criteria to assess all markets for bubbles. I have periodically shown you these for stocks and the stock market.

What Was Shown in January Versus Now

I will first describe the picture in words, then show it in charts, and compare it to the last update in January.

As of January, the bubble indicator showed that a) the US equity market was in a moderate bubble, but not an extreme one (ie., 70 percent of way toward the highest bubble, which occurred in the late 1990s and late 1920s), and b) the emerging tech companies (ie. As well, the unprecedented flood of liquidity post-COVID financed other bubbly behavior (e.g. SPACs, IPO boom, big pickup in options activity), making things bubbly. I showed which stocks were in bubbles and created an index of those stocks, which I call “bubble stocks.”

Those bubble stocks have popped. They fell by a third last year, while the S&P 500 remained flat. In light of these and other market developments, it is not necessarily true that now is a good time to buy emerging tech stocks.

The fact that they aren't at a bubble extreme doesn't mean they are safe or that it's a good time to get long. Our metrics still show that US stocks are overvalued. Once popped, bubbles tend to overcorrect to the downside rather than settle at “normal” prices.

The following charts paint the picture. The first shows the US equity market bubble gauge/indicator going back to 1900, currently at the 40% percentile. The charts also zoom in on the gauge in recent years, as well as the late 1920s and late 1990s bubbles (during both of these cases the gauge reached 100 percent ).

The chart below depicts the average bubble gauge for the most bubbly companies in 2020. Those readings are down significantly.

The charts below compare the performance of a basket of emerging tech bubble stocks to the S&P 500. Prices have fallen noticeably, giving up most of their post-COVID gains.

The following charts show the price action of the bubble slice today and in the 1920s and 1990s. These charts show the same market dynamics and two key indicators. These are just two examples of how a lot of debt financing stock ownership coupled with a tightening typically leads to a bubble popping.

Everything driving the bubbles in this market segment is classic—the same drivers that drove the 1920s bubble and the 1990s bubble. For instance, in the last couple months, it was how tightening can act to prick the bubble. Review this case study of the 1920s stock bubble (starting on page 49) from my book Principles for Navigating Big Debt Crises to grasp these dynamics.

The following charts show the components of the US stock market bubble gauge. Since this is a proprietary indicator, I will only show you some of the sub-aggregate readings and some indicators.

Each of these six influences is measured using a number of stats. This is how I approach the stock market. These gauges are combined into aggregate indices by security and then for the market as a whole. The table below shows the current readings of these US equity market indicators. It compares current conditions for US equities to historical conditions. These readings suggest that we’re out of a bubble.

1. How High Are Prices Relatively?

This price gauge for US equities is currently around the 50th percentile.

2. Is price reduction unsustainable?

This measure calculates the earnings growth rate required to outperform bonds. This is calculated by adding up the readings of individual securities. This indicator is currently near the 60th percentile for the overall market, higher than some of our other readings. Profit growth discounted in stocks remains high.

Even more so in the US software sector. Analysts' earnings growth expectations for this sector have slowed, but remain high historically. P/Es have reversed COVID gains but remain high historical.

3. How many new buyers (i.e., non-existing buyers) entered the market?

Expansion of new entrants is often indicative of a bubble. According to historical accounts, this was true in the 1990s equity bubble and the 1929 bubble (though our data for this and other gauges doesn't go back that far). A flood of new retail investors into popular stocks, which by other measures appeared to be in a bubble, pushed this gauge above the 90% mark in 2020. The pace of retail activity in the markets has recently slowed to pre-COVID levels.

4. How Broadly Bullish Is Sentiment?

The more people who have invested, the less resources they have to keep investing, and the more likely they are to sell. Market sentiment is now significantly negative.

5. Are Purchases Being Financed by High Leverage?

Leveraged purchases weaken the buying foundation and expose it to forced selling in a downturn. The leverage gauge, which considers option positions as a form of leverage, is now around the 50% mark.

6. To What Extent Have Buyers Made Exceptionally Extended Forward Purchases?

Looking at future purchases can help assess whether expectations have become overly optimistic. This indicator is particularly useful in commodity and real estate markets, where forward purchases are most obvious. In the equity markets, I look at indicators like capital expenditure, or how much businesses (and governments) invest in infrastructure, factories, etc. It reflects whether businesses are projecting future demand growth. Like other gauges, this one is at the 40th percentile.

Strange and complex behavior of frog cell blobs

A xenobot “parent,” shaped like a hungry Pac-Man (shown in red false color), created an “offspring” xenobot (green sphere) by gathering loose frog cells in its opening.

Tiny “living machines” made of frog cells can make copies of themselves. This newly discovered renewal mechanism may help create self-renewing biological machines.

According to Kirstin Petersen, an electrical and computer engineer at Cornell University who studies groups of robots, “this is an extremely exciting breakthrough.” She says self-replicating robots are a big step toward human-free systems.

Researchers described the behavior of xenobots earlier this year (SN: 3/31/21). Small clumps of skin stem cells from frog embryos knitted themselves into small spheres and started moving. Cilia, or cellular extensions, powered the xenobots around their lab dishes.

The findings are published in the Proceedings of the National Academy of Sciences on Dec. 7. The xenobots can gather loose frog cells into spheres, which then form xenobots.
The researchers call this type of movement-induced reproduction kinematic self-replication. The study's coauthor, Douglas Blackiston of Tufts University in Medford, Massachusetts, and Harvard University, says this is typical. For example, sexual reproduction requires parental sperm and egg cells. Sometimes cells split or budded off from a parent.

“This is unique,” Blackiston says. These xenobots “find loose parts in the environment and cobble them together.” This second generation of xenobots can move like their parents, Blackiston says.
The researchers discovered that spheroid xenobots could only produce one more generation before dying out. The original xenobots' shape was predicted by an artificial intelligence program, allowing for four generations of replication.

A C shape, like an openmouthed Pac-Man, was predicted to be a more efficient progenitor. When improved xenobots were let loose in a dish, they began scooping up loose cells into their gaping “mouths,” forming more sphere-shaped bots (see image below). As many as 50 cells clumped together in the opening of a parent to form a mobile offspring. A xenobot is made up of 4,000–6,000 frog cells.

Petersen likes the Xenobots' small size. “The fact that they were able to do this at such a small scale just makes it even better,” she says. Miniature xenobots could sculpt tissues for implantation or deliver therapeutics inside the body.

Beyond the xenobots' potential jobs, the research advances an important science, says study coauthor and Tufts developmental biologist Michael Levin. The science of anticipating and controlling the outcomes of complex systems, he says.

“No one could have predicted this,” Levin says. “They regularly surprise us.” Researchers can use xenobots to test the unexpected. “This is about advancing the science of being less surprised,” Levin says.