Description:
The assessment for this module will consist of two components:
(a) a single research project in an area of applied AI (written up in the form of a 3,000 word report) and
(b) a short 1-minute recorded presentation of the project. Topics for the research project will be based on one of the areas we will cover this semester, including feedforward neural networks, natural language processing or computer vision. Specifically, the ideas is to choose one of the areas below and extend a lab session into a full project + report.
Please note - you mat not be taught all aspects relating to your topic explicitly in class. It is expected that you will do further reading and research and find out what you need in terms of theoretical background or code base.
Topics are as follows:
1. Feedforward neural nets with hyperparameter optimisation: this project will implement a feedforward neural network for a dataset of choice and experiment systematically with a number of hyperparameter configurations,
e.g. the learning rate, batch size, number of hidden units, layers, etc. The project will need to explore systematic approaches for hyperparameter optimisation such as random or grid optimisation or genetic algorithms. Note that the specifics will not be taught directly, you’ll need to research and find out how to implement these in Python yourself. You may use any software libraries available, as long as referenced. The approaches named above e.g. come with sk_learn and do not need to be implemented from scratch. The neural network’s performance should be evaluated in different settings and compared against other approaches, such as decision trees, Naive Bayes or other classifiers. Results should be supported with visualisations, such as graphs.
2. Text classification: this project will implement a deep learning system for text classification (e.g. using the news dataset from the lab, or any other you can find). You can choose what classes you want to learn (i.e. classify) from your dataset. You will need to make an informed choice of neural network (such as recurrent or transformer) and implement it using a deep learning library. This part can be based on a lab session we did together. The project should also include at least one additional component, e.g. a specific hypothesis you want to investigate, a comparison against another technique, or a data augmentation technique, such as language modelling (i.e. embed features into vector space using one or more techniques e.g. Word2Vec, GloVe, BERT, GPT-2, etc.). Regardless of what you choose to do concretely, make sure that you use baselines in your project, i.e. choose the system you want to “pitch” and make sure you compare it another setup. Results should be supported with visualisations, such as graphs.
3. Sentiment analysis from text and/or images: this project will implement a deep learningbased system for sentiment analysis. You will need to choose a dataset (e.g. not the one used in our sentiment analysis lab please) and make an informed choice of architecture. Then implement it using a deep learning library. You can then either focus on sentiment analysis from text (as we’ve done before) or image analysis, e.g. predicting sentiment from images. In either case, please make sure to benchmark your results against an alternative setting, e.g. experimenting with more than one neural network architecture, or experimenting substantially with your chosen architecture itself, e.g. using hyperparameter optimisation. Results should be supported with visualisations, such as graphs.
Report Details
Introduction - an introduction to the topic, NOT to your report. Present your topic in the context of the field of AI, why is your topic important, why does it matter? What is your main research question? What is the expected outcome?
You can also prepare your readers for the rest of the report here, “Section 2 will introduce related work, Section 3… etc.” but this is often boring and might take words away from more important things.
Background - introduce related work to your project, i.e. the context in which your research should be seen and interpreted. What related work does already exist? This will require some background reading and literature review. Don’t just describe what research already exists, but discuss it in relation to your project - what is similar, what is different?
How does the related work link with your project? Are you aiming for an alternative method, an extension? a new dataset or application?
Objectives - state concisely your research objective/s. These need to be SMART - specific, measurable, attainable, realistic and time-bound. Don’t choose anything you couldn’t achieve within the time frame, but also be ambitious - don’t just replicate an online blog
Methodology - introduce your methodology from a technical but high-level point of view. You can use equations here or choose to describe your methodology (still needs to be concise, clean and technical though). Provide references to the model you have chosen for your project.
DO NOT include programming code into the report, i.e. screenshots or similar. If you want to present an algorithm, neural network architecture etc., then use pseudocode, a diagram or some other presentation that is not code copy-pasted code.
You may wish to include an architecture diagram of your approach or any other visual presentation. This normally helps the reader and makes the report look nicer
Experiments - Describe your experimental setup. What hyperparameters are you using for training? What dataset/s? What training-test split? What baselines, evaluation metrics?
Results - Present your results, ideally supported with tables and / or graphs. Discuss them, how do they compare with baselines? Did you meet your objectives? If not, why not? Did you find anything interesting, unexpected? Anything worth investigating further?
Conclusion - A brief section summarising the main points of your paper and findings. Make suggestions for future work - what experiments may follow from the work you did?
References - include a substantial number of relevant references. These should go beyond the literature resources provided for the module
In your marking criteria there is also a smaller rubric “quality of presentation” - this refers to the overall structure of the document, level of proofreading, and general presentation. It should be an easy section to get full marks on.
Presentation details: Your presentation should be a short and concise 1-minute pitch of your project idea. This is deliberately short to encourage you to focus on the main detail. You could structure your presentation around the format of an “elevator pitch”, see examples and ideas under these links: https://www.mmu.ac.uk/media/mmuacuk/content/documents/research/Impact-Tool----Elevator-Pitch.pdf https://graduateschool.nd.edu/assets/76988/elevator_pitch_8_28_2012.pdf https://medschool.vanderbilt.edu/wp-content/uploads/sites/9/files/public_files/ Elevator%20Pitches%20for%20Scientists_Uyen_0.pdf https://versatilehumanists.duke.edu/2018/10/23/crafting-an-academic-elevatorspeech-that-stands-out/
Please keep to the time of 1 minute. I’m not expecting any results in this as your project will still be under investigation/ development. You can support your presentation with a clean slide (recommended), and it should be pre-recorded and uploaded to Canvas.
To get solution of above problem you can comment in below comment section or send your requirement details at:
realcode4you@gmail.com
We are providing complete support to write research paper with code implementation with an fair price compare to other.
Comentários