Matt Scholar Solutions

Skills · Data science and AI

Deep learning for dissertation research

Learn to build and evaluate deep learning models for images, text and audio, using TensorFlow or PyTorch. You'll learn to fine-tune pretrained models on the compute you actually have, prevent overfitting on small datasets, and report results that are fair and reproducible.

Introduction

Learn to build and evaluate deep learning models for images, text and audio, using TensorFlow or PyTorch. You'll learn to fine-tune pretrained models on the compute you actually have, prevent overfitting on small datasets, and report results that are fair and reproducible.

When you need this skill

  • Your dissertation works with images, text, audio or other unstructured data.
  • You're fine-tuning a pretrained model, such as a vision model or a language model.
  • Your model overfits, or training doesn't converge.
  • You need to compare deep learning models fairly and reproducibly.

Common student questions

When should I use deep learning instead of classical machine learning?

Use deep learning for unstructured data, such as images, text and audio, where it learns useful features directly from raw inputs. For tabular data, classical methods are often better: Grinsztajn, Oyallon and Varoquaux (2022) found tree-based models still outperformed deep learning on medium-sized tabular datasets. Choose by your data type and dataset size, not by which method sounds more advanced.

What is transfer learning, and why should I use it?

Transfer learning starts from a model already trained on a large dataset, such as ResNet for images (He et al., 2016) or BERT for text (Devlin et al., 2019), then fine-tunes it on your smaller dataset. It usually needs far less data and compute than training from scratch, and often performs better, which makes it the realistic choice for most dissertations.

How do I prevent overfitting on a small dataset?

Start from a pretrained model, and use data augmentation to create realistic variations of your training examples. Add regularisation, such as dropout (Srivastava et al., 2014) and weight decay, and use early stopping based on validation performance. Monitor training and validation curves: if training performance keeps improving while validation performance falls, the model is overfitting.

Should I use TensorFlow or PyTorch?

Both are mature and widely used, and either is fine for a dissertation. PyTorch is common in research code, so it helps when you build on a published method's code. TensorFlow with its Keras interface offers a simple, high-level way to build standard models. Choose the framework your supervisor uses or the papers you build on use.

How do I make deep learning results reproducible?

Set random seeds for every library you use, record library versions and hardware, and save your configuration and model checkpoints. Some GPU operations are not fully deterministic, so results can still vary slightly. Train each model several times and report the mean and standard deviation, not your single best run.

How do I compare deep learning models fairly?

Use the same data splits, preprocessing and evaluation metrics for every model, and give each a similar tuning effort and compute budget. Always include a simple baseline. Report results across several runs, and use statistical tests where differences are small; our computer science subject page explains which tests to use.

What you'll learn

  • Choose the right architecture for your data: convolutional networks, transformers or sequence models
  • Load, preprocess and augment image, text or audio data
  • Fine-tune pretrained models with transfer learning
  • Train with sensible settings for learning rate, batch size, optimiser and epochs
  • Prevent overfitting with augmentation, regularisation and early stopping
  • Read training curves and diagnose common problems
  • Make experiments reproducible and track them
  • Evaluate, compare and explain models, including visual explanations for images

Tools required

Deep learning tools and their uses
ToolUsed for
PyTorchBuilding and training models, especially when using research code
TensorFlow with KerasBuilding and training models with a high-level interface
Google ColabTraining on a free cloud GPU, within its usage limits
Jupyter notebooksDocumenting experiments

How sessions work

Live online sessions, scheduled around your time zone. We build and fine-tune a model together on a sample public dataset, step by step, then you apply each step to your own project between sessions. Recordings and notebooks are available to revisit. This skill is taught in the Python and Machine Learning for Research.

Worked example

Sample project: Classifying plant leaf images as healthy or diseased, using a public image dataset and a free cloud GPU.

  1. 1. Data

    Check the class balance and image quality, and split by source so near-duplicate images don't appear in both training and test sets.

  2. 2. Baseline

    A small convolutional network trained from scratch.

  3. 3. Transfer learning

    Fine-tune a pretrained ResNet, first training only the final layers, then unfreezing more.

  4. 4. Overfitting control

    Augmentation with flips, rotations and colour changes, plus early stopping.

  5. 5. Evaluation

    Precision, recall and F1 per class, and a confusion matrix, across several runs with different seeds.

  6. 6. Explanation

    Review visual heatmaps to check that the model looks at the leaf, not the background.

  7. 7. Report

    The baseline versus the fine-tuned model, as mean and standard deviation across runs.

Practical guidance

FAQs

Do I need a powerful GPU?
Usually not. Fine-tuning a pretrained model on a free cloud GPU is often enough for a dissertation. Plan your compute when you scope the project.
Can I use a large language model in my dissertation?
Yes, as a subject of study, for example evaluating or fine-tuning one on a defined task. Check your university's rules on generative AI, and on what data you may send to online services.
Do I need to know machine learning first?
It helps. Data splitting, evaluation and avoiding leakage apply in the same way, and we cover them as needed.
My model's accuracy is very high. Should I worry?
Check for leakage, such as near-duplicate images across splits, and whether the model relies on shortcuts like backgrounds or watermarks. Visual explanations help reveal this.
Free 20-minute consultation · No obligation

Learn to build and fine-tune your model with a mentor

We will understand your course, stage, and specific roadblocks, and recommend a structured skills programme—or tell you honestly if we are not the right fit.

Discuss topic & timeline
Honest feasibility check
Pricing discussed transparently

Pricing is discussed in your free consultation.

Written by the Matt Scholar research team · Last reviewed 15 November 2026

References

  • Devlin, J., Chang, M.-W., Lee, K. and Toutanova, K. (2019) 'BERT: pre-training of deep bidirectional transformers for language understanding', Proceedings of NAACL-HLT 2019, pp. 4171–4186.
  • Grinsztajn, L., Oyallon, E. and Varoquaux (2022) 'Why do tree-based models still outperform deep learning on typical tabular data?', Advances in Neural Information Processing Systems 35 (Datasets and Benchmarks Track).
  • He, K., Zhang, X., Ren, S. and Sun, J. (2016) 'Deep residual learning for image recognition', Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 770–778.
  • Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I. and Salakhutdinov, R. (2014) 'Dropout: a simple way to prevent neural networks from overfitting', Journal of Machine Learning Research, 15, pp. 1929–1958.