FinTech Innovation and Cryptocurrencies Assignment Review

This set of vocabulary flashcards covers key concepts from the FinTech Innovation and Cryptocurrencies assignment, including machine learning architectures, GPT functionality, prompting techniques, and financial market theories.

Multi-Layer Perceptron (MLP)

A type of neural network consisting of multiple layers of nodes and hidden layers used in Transformer models to extract complex features and patterns from input data.

Temperature (GPT Parameter)

A parameter that controls the randomness of model output; higher values result in more creative and random text, while lower values make the output more focused and deterministic.

Generative Pre-trained Transformer (GPT)

A type of deep learning model that utilizes neural networks and an attention mechanism to perform tasks such as text generation.

Attention Mechanism

A critical architecture element in GPT models that prevents tokens from being represented by fixed embedding vectors and allows the model to compute scores between different tokens.

Role Prompting

A prompting technique where the user instructs the AI to act as a specific persona, such as a professor, to improve the quality or style of the response.

Negative Prompting

A technique used to instruct a model on what to avoid in its response, such as avoiding technical jargon, analogies, or mathematical formulas.

Safeguards Mechanism

Systems integrated into ChatGPT to prevent the generation of harmful or biased content, though they may occasionally produce false positives.

Retrieval-Augmented Generation (RAG)

A technique that combines pre-trained language models with external knowledge sources to improve the accuracy and relevance of the generated output.

Hallucination

A phenomenon in GPT models where the AI generates incorrect or nonsensical information, potentially caused by the complexity of the MLP architecture or limitations in the attention mechanism.

Algorithmic Trading

A method of executing trades at high speeds and frequencies using pre-defined rules or machine learning, which helps avoid human emotional biases and allows for $24/7$ market monitoring.

Efficient Market Hypothesis (EMH)

A theory stating that market efficiency depends on information availability and dissemination; it suggests that if a market is efficient, prices already reflect all relevant information.

Weak-Form Efficiency

A level of market efficiency where all historical price and volume data is reflected in stock prices, implying that machine learning algorithms cannot outperform the market using historical data alone.

Fundamental Analysis

The study of a company's intrinsic value based on financial and economic factors to make investment decisions.

Technical Analysis

The analysis of historical price patterns and trends to identify future trading opportunities.

Knowledge Cutoff Date

The point in time up to which a traditional GPT model was trained, after which it lacks information about world events unless supplemented by websearch or RAG.

Input Layer Nodes (Image Classification)

In a vanilla neural network, the number of nodes in the first layer equals the total number of pixels in the input image; for a $5 imes 5$ pixel image, there are $25$ nodes.

Output Layer Nodes (Digit Recognition)

In a neural network designed to classify handwritten digits ($0$-$9$), there are typically $10$ nodes in the output layer.