the course is machine learning. you are going to do part 3 and you should be based on part 2 answer refer "part 2 project answer (my work)" file and follow the steps and submit all the required tasks. to make it clear part 1 and 2 files i attached are the work i did in the past. the second file i attached, part 1 pdf is the proposal i did. third file i attached, "part 2 question pdf" is just part 2 question, fourth file i attached, "part 2 project answer (my work)" is the answer for part 2 question and you will be based on this file to do part 3.
part 3 project rev This is part 3 project and you should be based on part 1 and 2 project i did in the past and i attached my work. part 1 is just a proposal of a topics i proposed and on part 2 i chose only one topic of the two's topic i proposed in part 1 project and i chose topic 2 and worked on that topic, which is Autonomous Driving using Convolutional Neural Networks. on part 3 you will be working on this topic and see the work i did on "part 2 project answer (my work)" file, i attached it and you should be based on that. The attached "part 2 question" file is just the project question of part 2 in order for you to understand how i did the part 2 project so that you can do part 3 projects based on part 2 answer. Please follow the steps and answer all the tasks. It should includes all the required tasks graphs, tables, charts, datasets, figures, source codes………etc. use only academic library, library.csun.edu or google scholar. Part 3: compare your study with other algorithms and write your final report (20 points) Purposes: ● To identify other related tools that could be used in the same application. ● To recognize arguments, claims and strengths of previous work. ● Compare your results with findings of other papers and verify your assumptions or hypothesis. ● Justify your contributions (selling points of your work ) ! i.e. how are your results better than previous work ? e.g. your results generate a better test accuracy on the same dataset., your dataset uses fewer number of features to achieve the same test accuracy as previous work,…etc. ● Practice writing and finishing up results of academic project experiments. ● Practice how choose essential concepts, and ideas and remove unnecessary information within limitations of time. Tasks : Step1 : Beside the algorithm you chose to apply on part 2, choose another two or more similar algorithms to run comparisons. I attached part 2 file ● Run these algorithms on the same dataset that you used for part 2 and compare their accuracies and performance. Ex, compare their train, validation and test accuracies. ● Show and compare these results in graphs, charts or tables. ● Expand the "Results" section from part 2 to include these findings. Step 2: Compare pros and cons of these algorithms and include your findings in Discussion section. ● compare results, limitations, strengths of previous work with yours. Do your results support or oppose your original hypothesis ? Are your results better than previous work ? Step 3: Write up "abstract", " discussion", " limitations\future direction" and "conclusion" sections and expand the contents of part 2. ● The final draft should be 5~6 pages long in CSCSU format. You can check " Good student examples - A grades" as references. I attached it ● Here are details about each section: ○ “Abstract”: This section should describe the whole project briefly in a paragraph within 500 words. This serves as an attractive point to attract a reader attention. ○ Future work and limitation: what are limitations of your work ? Ex: the dataset is small and you plan to scale it up in the future; your study only focus on people in the States and you plan to expand study to include other countries …etc. ○ Conclusions: describe the results briefly again with selling points of your work. ○ Step 4: Include the draft, dataset , source code, all graphs, figures, tables into a zipped folder ● Our TA will verify your results so your results should match up what you wrote on the draft. Failing to run the simulation would be considered as cheating. ● Academic Integrity is taken seriously so do not copy or cite the whole paragraph from other papers. ● Quotations from other papers should not be over one sentence. Your paper will be checked by plagiarism software such as Turnitin. The similarity rate should be less than 20 %. Step 5: Upload the 5-pages draft via the submission of draft link for plagiarism checking. Step 6: A team would fail the project and lose 30 final points if the zipped folder fails to include the 5-page draft, source codes, dataset, simulating results and original files of figures/pictures/charts/tables used on the draft. Criteria for Success: ● You write findings of applying these different tools and listed pros and cons based on these findings. ● You compare your results with results found in previous work and list the pros and cons. You provide explanations and possible reasons to justify your hypothesis and results. You could have arguments from previous papers to support these justifications. ● You expand your work from part 2 and your final report should include sections as: ○ Abstract ○ Introduction ○ Datasets ○ Methodologies ■ Ex: data preprocessing process, models, tools, algorithms,..etc. ○ Results ○ Discussions ○ Limitations/ future work ○ Conclusions ○ Reference list ○ ● Your contents, citations and references satisfied all the format requirements and is in CSCSU format. ● Your reference list has at least 15 most related papers that are appropriately cited. ● Your paper has figures, graphs, tables or charts that make audience very easy to understand your results and they are in black and white colors. ● Your report does not have any typo or grammatical errors. ● Your report should include at least 5 pages of content long in the CSCSU format at most 6 pages including the page of reference list. ● Your report does not violate any copyright or have any plagiarism problems. All pictures, charts and graphs that are obtained from previous work should be cited and reference appropriately or have permissions from owners (e.g. authors or publishers who own the publication). part 2 question Part 2: finalize a project topic from results of part 1, select a suitable tool, find a small size of dataset and produce a preliminary result. (15 %). Purposes: -To identify the related tools. -To learn how to use a new library, an existing tool, or develop tools by your own and solve problems of installing software -To be able to find a small dataset (at least 30 subjects) and produce preliminary results. -To compare each tool with their pros and cons with different size of datasets. -To measure efficiency of each tool and record the results. -To finalize the project topic if previous topic does not work or cannot find related tool. -The finalized combination of topic and tool should be different from ones reported on other papers. These would be counted as your contributions. In other words, if some paper has used ID3 tree on weather data, you CANNOT use ID3 tree on weather again. Tasks: around 20 hours workload could be divided between team members - Step 1 : Run simulations on the combinations of tools/datasets that you submitted for part1 . ( Time hours: at least 6 ~ 8 hours ) . - Step 2: Choose the most feasible topic from two topics based on the results of previous step. ( Time required: depending on the level of success).In other words, you need to be able to run algorithms and tools on the datasets. Ex, you choose one deep learning algorithm and you cannot run it on your computer or it crashes after running two days. You might need to chose another topic. If you run out of two topics, then you might need to repeat what you have done before to search another suitable topic. Here are other things you need to consider for finalizing your topic: ○ Could you find related public datasets for you to use ? For example, could you find related public weather data in Los Angeles ? could you find public crime rate ? If you can not find related public data, you might need to choose another application field or even project topic. ○ If you cannot find public dataset, can you create questionnaire to collect data from other people ( must have more than 30 instances, as many as possible) ? ○ If you cannot find public tools, can you email authors for possible tools ? - Step 3: check if you dataset have more than 30 instances. ( Time required: 0.5 hours) A decent dataset should have 100 ~200 samples. If you cannot find enough samples, you might need to repeat step 1 &2 until you can have enough instances. - Step 4: Read more details about those 15 papers you have chosen for part 1 before and study their pros, cons, limitations of their research. Brainstorm with your partner about new perspective you could contribute to the project and write them down. (Time required: 2 ~ 3 hours ). ○ New perspective is not limited to new invention. Here are possible contributions to the projects. Ex, you could apply common algorithms on datasets in another field. ○ Ex: Contribute by providing additional data preprocessing and results show a better test accuracy ○ Ex: Very few people have applied algorithms on falling type of seniors so you contribute by conducting the experiments in a new field. ○ Ex: You optimize learning models ○ Ex: You find most influential features from a better feature selection process. ○ Ex: You decide and propose a better feature selection process...etc. - Step 5: Once finalizing a topic, run the work and generate preliminary work to verify your hypothesis. ( Time required; 1~2 hours) Then, if a team has fewer than 100 samples, one’s team could test on scope of the project by repeating those 30 subjects up to create 300, 3000 or more simulating data and run the tool. Estimate how much time it takes to run these data. - Step 6: Report your work in a MS word draft ( at least 3 pages ) ( Time required: 3~4 hours) in a CSCSU format that is posted under " Resources for doing project". You can check " Good student examples - A grades" under "Resources of doing project" module as references. You can also find average and bad examples there. - Step 7: Make sure all figures and tables should be created in black & white color. (Time required: 1~2 hours) - Step 8: Submit the draft via Canvas for plagiarism checking. The similarity rate checked by Turnitin should not be over 20%. Criteria for Success: your draft must contain 6 sections (1) Introduction - Describe a problem that one’s team plans to solve and give a brief background of this problem. - List themes (i.e. commonalities ) of related papers and describe what they have done to solve these problems. - Describe predictive models of these papers briefly. - List debatable places or limitations - Justify the reasons of your plan, and hypothesis . Why is your work important ? what has not be done by previous