search

TengMCing / bushfire-paper

Australian 2019-2020 bushfire season research
GNU General Public License v3.0
0 stars 0 forks source link

Thesis proposal structure #2

Closed dicook closed 3 years ago

dicook commented 4 years ago
TengMCing commented 4 years ago

Research plan

I have a question about the "research plan" section of the outline you suggested.

When I looked at the ETC4860 unit outline, it has an explanation of the research plan. It seems like it is a combination of data source and methodology for each part. Therefore, I wrote potential methods and issues to conduct this whole research on Thursday night.

However, given that you said the statement of topic including the research plan needed to be precise, and sections, later on, will also cover preliminary results, do I still need to give the whole methodology in the "research plan" section? Or what level of details should I provide in the "research plan" section? image

Scope of the research: what's feasible, what's not

Another question is about the section "Scope of the research: what's feasible, what's not".

I highly suspect that predict the fire ignition is infeasible, and we only can provide an ambiguous answer.

It's mainly because what we are actually doing is labelling the clusters in the unsupervised learning framework. The ideal situation will be only a few clusters been found and we successfully label them using our prior knowledge. But it also could be situations like clusters merge together that we will not be able to identify the label at all.

semi-supervised learning

It leads me thinking about another approach when I am writing this post, that perhaps we can turn it into semi-supervised learning. We don't have any label data from any open source, but we can create it by ourselves. We still have reports and articles about fire ignitions during 2019-2020 bushfire season and before the bushfire season. So we can manually label some of the fire. However, apply this method will be time-consuming for me, due to lack of knowledge in semi-supervised learning. Therefore, it introduces another potential infeasibility in our research.

There is another concern that if I decide to use this method, I need to also mention it in the literature review, research plan, model and timeline. I will be able to do that, but it will definitely put more pressure on my already tight schedule given I need to read more articles in a foreign field to write my literature review.

What's your thought? Should I include this new idea in my research proposal?

dicook commented 4 years ago

Hi Weihao,

My recommendation is to work on the thesis version. The shorter report can be made from parts of your thesis. This way you are on your way to having 1/3-1/2 of your thesis done.

predict the fire ignition is infeasible,

Yes, in the sense of we don't have labelled data, so we are proposing a risk factor based on other information.

we can turn it into semi-supervised learning

you are already doing semi-supervised learning. its a big label that means many things.

On 9 May 2020, at 3:45 am, Weihao Li notifications@github.com wrote:

Research plan

I have a question about the "research plan" section of the outline you suggested.

When I looked at the ETC4860 unit outline, it has an explanation of the research plan. It seems like it is a combination of data source and methodology for each part. Therefore, I wrote potential methods and issues to conduct this whole research on Thursday night.

However, given that you said the statement of topic including the research plan needed to be precise, and sections, later on, will also cover preliminary results, do I still need to give the whole methodology in the "research plan" section? Or what level of details should I provide in the "research plan" section? https://user-images.githubusercontent.com/45957646/81404104-1b661780-9178-11ea-8b47-bfdd42ab66bb.png Scope of the research: what's feasible, what's not

Another question is about the section "Scope of the research: what's feasible, what's not".

I highly suspect that predict the fire ignition is infeasible, and we only can provide an ambiguous answer.

It's mainly because what we are actually doing is labelling the clusters in the unsupervised learning framework. The ideal situation will be only a few clusters been found and we successfully label them using our prior knowledge. But it also could be situations like clusters merge together that we will not be able to identify the label at all.

semi-supervised learning

It leads me thinking about another approach when I am writing this post, that perhaps we can turn it into semi-supervised learning. We don't have any label data from any open source, but we can create it by ourselves. We still have reports and articles about fire ignitions during 2019-2020 bushfire season and before the bushfire season. So we can manually label some of the fire. However, apply this method will be time-consuming for me, due to lack of knowledge in semi-supervised learning. Therefore, it introduces another potential infeasibility in our research.

There is another concern that if I decide to use this method, I need to also mention it in the literature review, research plan, model and timeline. I will be able to do that, but it will definitely put more pressure on my already tight schedule given I need to read more articles in a foreign field to write my literature review.

What's your thought? Should I include this new idea in my research proposal?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/TengMCing/bushfire-paper/issues/2#issuecomment-625935195, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAB52B7YPUKMCT3ZP7OSEP3RQRALPANCNFSM4M3XOIZQ.

Di Cook visnut@gmail.com