Knowledge check

Quiz

Complete

Evaluate your understanding of this unit by completing the Knowledge Check.

A data science team is developing a model to predict customer churn. Initial models show high accuracy on the training data but perform poorly on new, unseen data. Which model architecture component and associated techniques would be MOST relevant to investigate and potentially adjust to address this issue?

A) Input and feature representation: Reviewing the raw data sources and ensuring all relevant features are included.B) Model parameters and structure: Increasing the number of layers in a neural network or the depth of a decision tree to capture more complex patterns.C) Regularisation and hyperparameters: Experimenting with L1 or L2 penalties, adjusting dropout rates, or implementing early stopping during training.D) Optimisation algorithm: Switching from stochastic gradient descent (SGD) to Adam to potentially improve convergence speed.

You are building a natural language processing model to classify customer reviews as positive or negative. You notice that the model performs well on longer reviews but struggles with very short, one-word reviews. Considering the model architecture components, which area would offer the MOST direct way to potentially improve the model's performance on these shorter inputs?

A) Loss (objective) function: Changing from binary cross-entropy to hinge loss to adjust the penalty for misclassified short reviews.B) Optimisation algorithm: Implementing a mini-batch gradient descent strategy to ensure the model sees a variety of short and long reviews during each training step.C) Inferences and prediction mechanisms: Adjusting the threshold for classifying a review as positive or negative based on the model's output probability.D) Input and feature representation: Revisiting the text vectorisation technique to ensure short reviews are adequately represented and potentially exploring techniques that are more sensitive to single-word context.

Your organisation is deploying a machine learning model as a web service to provide real-time product recommendations. The data science team needs to ensure the model is consistently receiving the latest product catalogue and user interaction data for accurate recommendations. Which aspect of the system architecture is MOST critical to establish and maintain for this purpose?

A) MLOps tools: Implementing tools for model versioning and rollback in case of deployment issues.B) Cloud platforms: Selecting a cloud provider with robust compute resources for hosting the web service.C) Containerisation: Packaging the model and its dependencies into a Docker container for consistent deployment.D) Data ingestion: Establishing reliable data pipelines to automatically pull the latest product catalogue and user interaction data from their respective sources into the system.

A small healthcare startup has developed a novel medical imaging device but has a very limited dataset of labelled images for a rare disease. They want to build a classification model to automatically detect this disease in new scans. Considering their limited data, which of the following model training methodologies would likely be the MOST effective starting point?

A) Training a large convolutional neural network from scratch on their limited dataset.B) Using unsupervised learning techniques like clustering to identify potential disease patterns.C) Employing transfer learning by using a pre-trained model on a large, general medical image dataset and fine-tuning it on their specific, small disease dataset.D) Implementing federated learning by collaborating with other hospitals to aggregate their small datasets.

A research team is developing an AI model to generate realistic 3D models of furniture based on text descriptions. They have access to a powerful computing cluster with many GPUs, and the training process for their initial models is taking several weeks. Which of the following strategies would be the MOST appropriate to significantly reduce the training time?

A) Applying early stopping to prevent overfitting.B) Implementing hyperparameter tuning using grid search.C) Utilising distributed training to split the model training workload across the available GPUs in their computing cluster.D) Switching to a simpler model architecture with fewer parameters.

A company has built a sentiment analysis model for customer reviews. However, they are concerned that subtle changes in language or the use of slang might cause the model to misclassify reviews. To improve the model's robustness against such variations, which of the following advanced training techniques would be the MOST suitable to implement?

A) Contrastive learning using pairs of reviews with similar sentiment but different phrasing.B) Reinforcement learning from human feedback (RLHF) by having human annotators rate the model's sentiment predictions.C) Adversarial training by generating slightly modified reviews that are designed to fool the current model and training the model to correctly classify these "adversarial" examples.D) Transfer learning from a language model trained on a different domain of text data.

Your organisation has provided you with a standard office laptop (CPU, 16GB RAM) and tasked you with training a model to classify customer reviews (a medium-sized text dataset). Initial experiments show that a complex Transformer model achieves the desired accuracy but runs very slowly and occasionally crashes due to memory issues. Considering the hardware constraints and the need for a more feasible approach, which of the following strategies would be the MOST appropriate first step to try?

A) Immediately request access to a cloud-based TPU instance to handle the model complexity.B) Implement data parallelism by splitting the dataset and training the model on multiple instances of your laptop simultaneously.C) Apply model compression techniques like pruning or quantisation to reduce the Transformer model's size and computational demands.D) Increase the effective batch size by using gradient accumulation during training.

A research team has access to a cluster of high-end GPUs with excellent interconnect bandwidth and is training a very large image segmentation model. During training, they notice that the GPUs are not being fully utilised, and the training process seems slower than expected despite the powerful hardware. Considering the principles of distributed training, which of the following is the MOST likely bottleneck they are encountering?

A) Insufficient CPU resources on the master node to manage the distributed training process.B) The model architecture is inherently sequential and cannot be easily parallelised across the GPUs.C) The batch size used for training is too small, not providing enough work for each GPU to process in parallel.D) Fault tolerance mechanisms like frequent checkpointing are consuming a significant portion of the training time.

Your organisation is becoming increasingly aware of the environmental impact of its machine learning operations. You are training large language models in the cloud and want to minimise the carbon footprint. Which of the following strategies would be the MOST direct way to reduce the environmental impact associated with your cloud-based training?

A) Implement mixed precision training to accelerate the training process and reduce the total energy consumption for each training run.B) Optimise your model architecture using Efficient Architecture Search (EAS) to find a smaller model with comparable performance, requiring less computation.C) Migrate your cloud-based training workloads to a cloud provider that sources a higher percentage of its energy from renewable sources.D) Increase the frequency of saving model checkpoints during training to ensure no progress is lost in case of interruptions.

A data science team is experimenting with different hardware configurations for training a large language model in the cloud. They run the same training job on two different virtual machine setups: one with older, less energy-efficient GPUs and another with newer, more energy-efficient GPUs. Both setups achieve similar training times. Which of the following outcomes is MOST likely regarding the environmental impact of these two training runs?

A.) The training run on the older GPUs will likely have a lower environmental impact because older hardware typically has lower power consumption.B) The training run on the newer GPUs will likely have a lower environmental impact because they consume less power for the same amount of computation.C) The environmental impact of both training runs will be approximately the same since the training times were similar.D) The environmental impact will depend solely on the cloud provider's renewable energy sources, making the hardware efficiency irrelevant.

Action item: Knowledge check

In progress