CS P3

Latency

The delay between a user's query and the chatbot's response. High latency can negatively impact user experience by making the chatbot seem slow

Linguistic nuances

the subtle differences or variations in meaning, tone, or expression within language

Architecture

underlying structure and components that enable it to understand and generate human

Natural Language Processing (NLP)

The field of AI focused on enabling machines to interpret understand and respond to human language by breaking down and analyzing text through methods like lexical analysis and semantic analysis.

Recurrent neural networks (RNNs)

Neural networks designed to handle sequential data with memory of previous inputs.

Consist of input, hidden, and output layers

Change using backpropagation through time(BPTT) (vanishing gradient problem making it harder to learn long-term dependencies)

Long short-term memory (LSTM)

Type of RNN that overcomes vanishing gradient problem (with specialised units for capturing long-term dependencies)

Uses a three gate mechanism (input gate, forget gate, and output gate) to selectively retain or forget information, allowing them to maintain a long-term dependencies

Transformer neural networks (transformer NNs).

Neural networks using a self-attention mechanism for parallel processing and better handling of long-term dependences

Uses a self attention mechanism to capture relationships between the words in a sequence, enabling better handling of long-term dependence in parallel processing of data

Self-attention mechanism

Mechanism used in transformer neural networks that allows the model to weigh the importance of different words in the input sequence when making predictions

Dataset

A collection of data used to train and evaluate machine learning models. A good dataset is diverse, high-quality, relevant and up-to-date

Synthetic data

Artificially generated data used to supplement real data, covering scenarios that may not be well-represented in the original dataset

Diversity

Inclusion of a wide range of topics, languages, and user intents in the dataset.

Data Cleaning

Removing irrelevant, duplicate, and noisy data to ensure the remaining data is accurate and well-labelled

Data Augmentation

Techniques used to increase the size and diversity of the dataset.

Bias Mitigation

Analyzing the dataset for potential biases and taking steps to address them, such as balancing the representation of different user groups and scenarios.

User Feedback

Incorporating feedback from users to identify and correct inaccuracies, continuously improving the dataset's quality.

Pre-processing

The initial step in data preparation, involving cleaning, transforming, and reducing data to improve its quality and make it suitable for training machine learning models

Ethical challenges

Data privacy and security

Bias and fairness

Accountability and responsibility

Transparency

Misinformation and manipulation

Data Privacy

Ensuring that user data is kept confidential and secure.

Bias

Systematic errors in data or algorithms that lead to unfair outcomes.

Accountability

Responsibility for the actions and decisions made by a system.

Transparency

Clarity and openness about how a system operates and makes decisions.

Misinformation

Incorrect or misleading information.

Improve Data Privacy and Security

Implement encryption and access controls to protect customer data handled by the chatbot.

Improve Bias and Fairness

Use diverse training data and regular bias audits to ensure the chatbot provides fair service to all customers.

Improve Accountability and Responsibility

Establish clear guidelines that define who is responsible for the chatbot's actions and ensuring adherence to ethical standards.

Improve Transparency

Provide explanations for the chatbot's responses and educating customers on how the chatbot works.

Improve Misinformation and Manipulation

Integrate fact-checking mechanisms to ensure the accuracy of the information provided by the chatbot

Critical Path

The shortest sequence of models required to process a query.

Critical path optimisation

A method used to improve response time by identifying and eliminating unnecessary processing steps, helping chatbots respond more quickly and efficiently.

Natural language understanding (NLU)

A component of NLP focused on understanding the user's input by analysing linguistic features and context.

Lexical analysis

1.The process of breaking down text into individual words and sentences, identifying parts of speech, and preparing it for further processing

Syntactic analysis (parsing)

2.Analysing the grammatical structure of a sentence, identifying the relationships between words and phrases

Semantic analysis

3.The process of understanding the meaning of words and sentences, going beyond the surface-level structure to interpret the underlying concepts

Discourse integration

4.A stage in NLP where the meaning of a sentence is integrated with the larger context of the conversation to generate coherent and contextually appropriate responses

Pragmatic analysis

5.Analysing the social, legal, and cultural context of a sentence to understand its intended meaning and implications.

User Satisfaction

A measure of how well a chatbot meets user needs and expectations, influenced by response speed, accuracy, and relevance to the user's emotional tone and query

Attention-retention

The ability of a chatbot to keep users engaged by maintaining relevance and providing timely, helpful responses, preventing users from losing interest

Pattern Recognition

The ability of chatbots to identify recurring language patterns, which helps improve their understanding of different linguistic structures and user intents

Hyperparameters

learning rate and the number of hidden layers

Hyperparameter Tuning

The process of optimizing the parameters that govern the training of a machine learning model (e.g., learning rate, number of layers) to improve its performance.

Backpropagation through time (BPTT)

A variant of the backpropagation algorithm used for training Recurrent Neural Networks (RNNs), where gradients are propagated backward through time to update the weights.

Vanishing Gradient

A problem in training deep neural networks where gradients become very small, making it difficult to update the weights effectively and learn long

Weights

Parameters in a neural network that are adjusted during training to minimize the loss function and improve the model's predictions.

Biases

Systematic errors in data or algorithms that can lead to unfair or discriminatory outcomes

Confirmation bias

A type of bias where data is skewed towards a particular viewpoint or expected outcome, often reinforcing pre

Historical bias

A bias that occurs when training data reflects outdated or historical patterns that may not be relevant to current scenarios, potentially leading to inaccurate predictions

Labelling bias

Occurs when the labels applied to training data are subjective, inaccurate, or incomplete, affecting the model's ability to learn correctly

Linguistic bias

Bias resulting from training data that favors certain dialects, vocabularies, or linguistic styles, potentially disadvantaging users who use different linguistic forms

Sampling bias

Occurs when the training dataset is not representative of the entire population, leading to a model that performs well for certain groups but poorly for others

Selection bias

Bias introduced when the training data is not randomly selected but chosen based on specific criteria, potentially missing important variations

Bag of words

A model used in natural language processing where text is represented as an unordered collection of words, disregarding grammar and word order but keeping track of word frequency

Processing power

the computational capacity of a system to perform tasks efficiently and quickly. crucial for handling the complex algorithms and large datasets required for natural language processing (NLP), machine learning, and generating real-time responses. ensures that a chatbot can function smoothly, provide quick responses, and handle a high volume of queries simultaneously

Central Processing Unit (CPU)

The primary component responsible for executing instructions. While capable, CPUs may struggle with the high parallel processing demands of advanced AI tasks.

Graphical processing units (GPUs)

Specialised hardware designed to accelerate the processing of large

Tensor processing units (TPUs)

Custom hardware developed by Google specifically designed to accelerate machine learning workloads, particularly for deep learning models.

Cloud Computing

Utilizing cloud services (e.g., AWS, Google Cloud, Azure) provides scalable resources that can be adjusted based on demand. This flexibility ensures that processing power can be scaled up or down as needed.

Distributed Computing

Distributing tasks across multiple machines to parallelize processing, reducing latency and improving efficiency.

Large language model (LLM)

Advanced neural networks trained on vast amounts of text data to understand and generate human

Memory cell state

In LSTM networks, the memory cell state represents the information that flows through the network, controlled by input, forget, and output gates to manage long

Loss function

A mathematical function that measures the difference between the predicted output of a model and the actual target output, guiding the optimization process during training.

Self-attention mechanism

Mechanism within neural networks that allows the network to weigh the importance of different input elements when making predictions or decisions

Deep Learning

A subset of machine learning involving neural networks with many layers (deep neural networks) that can learn complex patterns in large datasets.