tags : education, Computation and Computer Theory, Math, Cricket’24
How to read paper?
3 approaches
Solution Finding
Immediate solutions for problems at hand, similar to reading SO/reference books.
Discovery
This is basically going down the rabbithole idea but which a aim to improve/optimize find better alternatives to your original idea.
Curiosity
Just reading things out of curiosity! It can surprise you how useful these random ideas can be :)
Other tips
- Avoid reading new papers at all. Focus on the classics, and let time filter out papers that are worth reading.
- Paper writers are mostly one of these three: Academics, Engineers, Industrial Labs
- Identify the methods used. They are the primary metric when analyzing results.
- Long term learning is a good way to get time for reading research papers.
Papers/Editorials
Here’s the org-mode format of academic publication types with each section and its items organized in org-tables:
Primary Research Publications
| Type | Description |
|---|---|
| Original Research Articles | Reports new findings from empirical studies, experiments, or analyses |
| Brief Reports/Short Communications | Shorter articles presenting preliminary or limited findings warranting rapid publication |
| Case Studies/Case Reports | Detailed analyses of individual cases, common in medicine, psychology, and business |
Theoretical and Conceptual Works
| Type | Description |
|---|---|
| Theoretical Papers | Present new theories, frameworks, or conceptual models without necessarily including empirical testing |
| Methodological Papers | Introduce new research methods, statistical approaches, or analytical techniques |
Scholarly Communication Formats
| Type | Description |
|---|---|
| Conference Papers/Proceedings | Research presented at academic conferences, often published in collected volumes |
| Preprints | Early versions of research papers shared before formal peer review |
| Book Chapters | Contributions to edited volumes focusing on specialized topics |
| Monographs | Book-length studies on a single subject by a single author or author team |
Applied Research Outputs
| Type | Description |
|---|---|
| Technical Reports | Detailed reports on technical projects by government agencies, research institutions, or companies |
| White Papers | Authoritative reports that inform readers about complex issues and present an organization’s philosophy |
| Policy Briefs | Short documents specifically aimed at government or organizational decision-makers |
Educational and Outreach Publications
| Type | Description |
|---|---|
| Textbooks | Comprehensive educational resources on specific subjects |
| Popular Science Articles | Research communicated for non-specialist audiences |
Meta-Research and Commentary
| Type | Description |
|---|---|
| Commentaries/Opinion Pieces | Scholarly perspectives on current research, often invited by journals |
| Letters to the Editor | Brief communications responding to previously published articles |
| Editorials | Pieces written by journal editors highlighting important research or issues |
Doing Reviews
| Review Type | Purpose | Methodology | Time Required |
|---|---|---|---|
| Systematic Review | Answer a specific research question with minimal bias | Comprehensive search with explicit protocol, inclusion/exclusion criteria | High |
| Meta-Analysis | Quantitatively combine and analyze results from multiple studies | Statistical pooling of data from similar studies | High |
| Realist Review | Understand what works, for whom, in what contexts | Theory-driven approach examining mechanisms | High |
| Integrative Review | Synthesize findings from diverse methodologies | Incorporates experimental and non-experimental research | Medium to High |
| Scoping Review | Map key concepts and evidence in a research area | Broad search strategy without quality assessment | Medium |
| State-of-the-Art Review | Provide current understanding and future directions | Comprehensive search of recent advances | Medium |
| Umbrella Review | Synthesize multiple systematic reviews | Systematic search for published reviews only | Medium |
| Critical Review | Critically evaluate literature with deep analysis | In-depth analysis of a manageable number of works | Medium |
| Narrative Review | Provide a broad perspective on a topic | Selective literature search based on author expertise(subjective) | Low to Medium |
| Rapid Review | Provide timely evidence for urgent decisions | Streamlined systematic review methods | Low |
Doing Literature Review
- Define your research question clearly.
- Starting with a clear research question prevents feeling overwhelmed by the amount of available research. For example, when researching the “Bayesian brain,” a research question could be “What does the Bayesian brain mean? How does it work?“.
- Find a good review and accompanying relevant research
- For this we can also use https://elicit.com/
- When new to a topic, she searches for reviews (e.g., “Bayesian brain review”) and sorts them by highest citation to find key foundational papers
- She also identifies leading authors in the field using Scopus to understand who the major contributors are (e.g., Karl Friston in computational neuroscience).
- Import these papers into a “dedicated” folder in Zotero or similar thing
- Put the set of selected papers into citation mapping software
- Eg. Litmaps, ResearchRabbit etc.
- Get more sources
- Read through them / analyze
- Do classical reading/note-taking
- Put all of those sources into NotebookLM
- Usually you’d want to generate a TOC of each source(unchecking other sources) to understand your source
- Organize insights
- distilling each paper into a main question, a key statement, or a central fact
- Synthesize the literature into your own review.
- involves thinking critically about the acquired knowledge, interpreting it, identifying gaps, and formulating your own questions.
Using the literature review for knowledge work
- Now that we have GOOD inputs in our
notebookLMsource list- We can now “ask the question” we want to ask via
notebookLM. This is like a much better google search results.
- We can now “ask the question” we want to ask via
- After we have got a good enough response, we can put that to a more creative LLM like
claudenotebookLMis tuned to hallucinate less. So it works good with factual data.
AI Research in India
Deploying ML applications (applied ML) @novasarc01 i’ll tell what’s really happening:
-
there is no strong ecosystem of research here even in the so called top institutes…research is in hands of few institutions…and any local student who has ideas or wants to do the research can’t access it easily…the students in these institutions don’t have interest or motivation to pursue meaningful research projects like open source LLMs, mech interp, AI alignment, VLMs
-
also there are no specialized research groups and labs in institutes like RL group, Interpretability Group, Parallel Programming Group, Semiconductor research group, marine robotics group where students from any college of the country can take part and do research under profs…why no specialized research group? no support from the institutions
-
the condition of local state unis is even worse…forget about research folks are not even interested in learning specialized skills and their only goal is to get out of the uni instantly…you can’t even think to start specialized research groups here
-
percentage of people who are serious about their research is very low…and they go outside the country to pursue research (and they should definitely go) considering the opportunities, already established strong research ecosystem and SOTA infrastructure
-
and its not new…from past few years i too was wondering why there are no huge open source projects in india where people are contributing…even the top cream fails to build such projects
-
there is no support for students starting from high school to develop interest in research…when i used to present ideas for doing research projects people used to laugh at me and used to say that instead of doing homework i am giving excuse to do other things…i also contacted few researchers but there was no response
-
today we have unimaginable opportunities from other countries which include microgrants for research, research fellowships, research project groups…and i am proud to see that people who are really interested in pursuing research are already leveraging them