Introduction To Artificial Intelligence
Expert-defined terms from the Professional Certificate in Artificial Intelligence for Investment Management course at LearnUNI. Free to read, free to share, paired with a professional course.
A set of computer‑driven rules that automatically generate and place buy or sell… #
Example: a momentum‑based algorithm that buys stocks when their 5‑minute price increase exceeds 0.5 %. Practical application: reduces human latency and can exploit micro‑price movements. Challenges: model over‑fitting, market impact, regulatory scrutiny.
The broader scientific discipline that seeks to create systems capable of perfor… #
Example: an AI system that predicts credit risk using alternative data sources. Practical application: automates decision‑making in portfolio construction. Challenges: interpretability, bias, data privacy.
A computational model inspired by the brain’s network of neurons, consisting of… #
Example: a feed‑forward ANN that forecasts daily returns of a commodity. Practical application: captures non‑linear relationships in market data. Challenges: hyper‑parameter tuning, risk of over‑parameterization.
A learning algorithm that computes the gradient of the loss function with respec… #
Example: updating weights in a multilayer perceptron after each batch of price data. Practical application: enables training of deep models for asset‑price prediction. Challenges: vanishing/exploding gradients, computational cost.
An ensemble technique that builds multiple models on bootstrapped subsets of dat… #
Example: a bagged decision‑tree ensemble that predicts bond rating migrations. Practical application: reduces variance and improves out‑of‑sample performance. Challenges: diminishing returns with many trees, increased memory usage.
A statistical method that updates the probability of a hypothesis as more eviden… #
Example: updating the probability of a market regime shift after new macroeconomic data. Practical application: provides probabilistic forecasts for risk management. Challenges: choosing appropriate priors, computational intensity for high‑dimensional models.
A measure of a security’s sensitivity to movements in a benchmark index, often u… #
Example: a stock with β = 1.2 is expected to move 12 % when the market moves 10 %. Practical application: informs portfolio weighting and risk budgeting. Challenges: assumes linear relationship, may vary across market cycles.
A predictive model whose internal mechanics are not easily understandable by hum… #
Example: a deep‑learning model that predicts FX rates without transparent feature importance. Practical application: can achieve high accuracy when data patterns are intricate. Challenges: regulatory compliance, trust, difficulty diagnosing errors.
An ensemble technique that sequentially adds weak learners, each focusing on the… #
Example: XGBoost model that forecasts equity volatility using lagged technical indicators. Practical application: often yields state‑of‑the‑art performance on tabular finance data. Challenges: prone to over‑fitting if not regularized, sensitive to noisy data.
A set of unsupervised learning techniques that group observations based on simil… #
Example: clustering assets by return‑correlation patterns to identify sectoral groups. Practical application: aids in diversification and factor construction. Challenges: selecting number of clusters, handling high‑dimensionality.
A deep learning architecture that applies convolutional filters to capture spati… #
Example: using a CNN to analyse heat‑map representations of order‑book depth. Practical application: extracts localized patterns from time‑series heat maps. Challenges: requires large labeled datasets, may be over‑kill for simple numeric series.
A model‑evaluation technique that partitions data into multiple folds to assess… #
Example: 5‑fold cross‑validation of a random‑forest model predicting default probabilities. Practical application: provides robust error estimates before deployment. Challenges: time‑series data require forward‑chaining to preserve temporal order.
The inadvertent inclusion of information in the training set that would not be a… #
Example: using future price as a feature when training a model to predict intraday moves. Practical application: vigilance prevents inflated back‑test results. Challenges: subtle in complex pipelines, especially with lagged features.
A subset of machine learning that employs multi‑layered neural networks to autom… #
Example: a deep LSTM that captures long‑term dependencies in macroeconomic time series. Practical application: improves forecasting accuracy for non‑linear dynamics. Challenges: requires substantial computational resources and careful regularization.
Techniques that reduce the number of variables while preserving essential inform… #
Example: applying Principal Component Analysis to compress a 200‑factor risk model into 10 principal components. Practical application: mitigates curse of dimensionality and speeds up training. Challenges: loss of interpretability, selection of retained variance threshold.
The practice of combining multiple models to achieve superior predictive perform… #
Example: blending a gradient‑boosted tree, a support‑vector machine, and a neural network for equity selection. Practical application: diversifies model risk and stabilizes forecasts. Challenges: increased complexity, difficulty in model governance.
The process of creating, transforming, or selecting variables that improve model… #
Example: constructing a “rolling‑beta” feature from historic price returns. Practical application: adds domain knowledge to data‑driven models. Challenges: labor‑intensive, risk of introducing leakage, may become obsolete as market dynamics evolve.
A linear classification method that projects data onto a line maximizing class s… #
Example: distinguishing “up‑trend” versus “down‑trend” periods in a currency pair. Practical application: provides a simple, interpretable decision boundary. Challenges: assumes normally distributed classes, limited to linear separability.
Sequences of data points indexed in time, such as prices, volumes, or spreads, t… #
Example: daily closing prices of a S&P 500 constituent. Practical application: foundation for forecasting, risk modeling, and algorithmic strategy design. Challenges: non‑stationarity, noise, structural breaks.
A non‑parametric probabilistic model that defines a distribution over functions,… #
Example: using a GP with a Matérn kernel to model implied volatility surfaces. Practical application: provides predictive intervals useful for risk budgeting. Challenges: cubic computational scaling with data size, kernel selection.
An iterative optimization algorithm that updates model parameters in the directi… #
Example: training a logistic regression model for credit scoring. Practical application: core technique for fitting many AI models. Challenges: choice of step size, convergence to local minima, sensitivity to feature scaling.
The practice of taking offsetting positions to reduce exposure to adverse price… #
Example: using options to hedge a long equity position against market downturns. Practical application: protects portfolio value while maintaining upside potential. Challenges: cost of hedge, model risk in estimating hedge ratios.
The process of selecting the optimal configuration of model settings that are no… #
g., tree depth, regularization strength). Example: employing random search to find the best number of LSTM units for a macro‑forecasting model. Practical application: enhances model accuracy and robustness. Challenges: computational expense, risk of over‑fitting to validation set.
A situation where the distribution of target classes is heavily skewed, often le… #
Example: default events represent only 1 % of a loan portfolio. Practical application: requires resampling or cost‑sensitive methods to improve minority‑class detection. Challenges: performance metrics can be misleading, over‑sampling may introduce noise.
A performance metric that measures risk‑adjusted excess return relative to a ben… #
Example: an IR of 0.6 indicates that the strategy’s excess return is 0.6 % per unit of active risk. Practical application: used to evaluate portfolio manager skill. Challenges: sensitive to estimation window, assumes normality of returns.
Algorithms that implicitly map data into high‑dimensional feature spaces using k… #
Example: a Support Vector Machine with radial basis function kernel classifying market regimes. Practical application: powerful for small‑to‑medium sized datasets. Challenges: kernel choice, scaling to large datasets.
A measure of how one probability distribution diverges from a reference distribu… #
Example: evaluating the difference between predicted and actual return distributions. Practical application: used in model calibration and portfolio optimization. Challenges: asymmetry, undefined when reference probability is zero.
A type of recurrent neural network architecture designed to capture long‑range d… #
Example: an LSTM that predicts quarterly earnings surprises from past financial statements. Practical application: improves forecasting accuracy for sequential data. Challenges: high computational cost, risk of over‑fitting with limited data.
A stochastic process where the probability of moving to the next state depends o… #
Example: modeling credit rating migrations as a discrete‑time Markov chain. Practical application: provides transition matrices for risk models. Challenges: assumption of stationarity, may oversimplify real market dynamics.
A portfolio construction technique that seeks the optimal allocation by minimizi… #
Example: solving a quadratic program to allocate capital across equities, bonds, and commodities. Practical application: foundational for modern portfolio theory. Challenges: estimation error in inputs, sensitivity to outliers.
A learning paradigm where the algorithm learns how to learn, enabling rapid adap… #
Example: a meta‑learner that quickly calibrates a pricing model for a newly listed asset. Practical application: reduces the time required for model re‑training. Challenges: complex training procedures, risk of negative transfer.
A computational technique that uses repeated random sampling to estimate the pro… #
Example: simulating 10,000 paths of a portfolio’s value under stochastic interest rates. Practical application: assesses Value‑at‑Risk (VaR) and scenario analysis. Challenges: convergence speed, model risk from underlying assumptions.
A field of AI that focuses on the interaction between computers and human langua… #
Example: using BERT to extract sentiment from earnings call transcripts. Practical application: informs trading signals based on news flow. Challenges: domain‑specific jargon, data sparsity, model interpretability.
An automated process that discovers optimal neural network topologies for a give… #
Example: NAS that designs a lightweight CNN for real‑time option‑pricing inference. Practical application: reduces manual model‑design effort. Challenges: high computational demand, risk of over‑fitting to validation set.
Models that do not assume a fixed functional form for the underlying data distri… #
Example: a k‑nearest‑neighbors model that predicts bond spreads based on similar historical observations. Practical application: adapts to irregular data structures. Challenges: curse of dimensionality, sensitivity to noise.
A modeling error where a model captures noise instead of the underlying signal,… #
Example: a deep network that memorizes training set price patterns but fails on out‑of‑sample periods. Practical application: detection prompts model simplification. Challenges: balancing bias‑variance trade‑off, especially with limited financial data.
The process of allocating capital across assets to achieve a desired risk‑return… #
Example: a multi‑objective optimizer that simultaneously maximizes Sharpe ratio and minimizes turnover. Practical application: constructs efficient, diversified portfolios. Challenges: estimation error, transaction costs, dynamic market conditions.
The creation of statistical or machine‑learning models that forecast future valu… #
Example: a logistic regression that predicts probability of a stock’s price breaking a resistance level. Practical application: informs trade entry and exit decisions. Challenges: data snooping, model drift.
An approach that provides a full probability distribution of future outcomes rat… #
Example: generating a 95 % prediction interval for next week’s commodity price. Practical application: supports risk‑aware decision making. Challenges: calibration of uncertainty, computational overhead.
A discipline that applies mathematical models, statistical techniques, and compu… #
Example: using stochastic calculus to price exotic options. Practical application: underpins systematic investment strategies. Challenges: model risk, reliance on historical data, regulatory compliance.
A machine‑learning paradigm where an agent learns to make sequential decisions b… #
Example: an RL agent that learns optimal execution schedules to minimize market impact. Practical application: automates adaptive trading policies. Challenges: exploration‑exploitation balance, sample inefficiency, stability of training.
Techniques that add constraints or penalties to model parameters to prevent over… #
Example: applying Lasso regularization to a linear factor model to enforce sparsity. Practical application: improves out‑of‑sample performance. Challenges: selecting appropriate regularization strength, potential bias introduction.
A portfolio construction method that allocates capital such that each asset cont… #
Example: a risk‑parity fund that balances equities, bonds, and commodities based on their volatilities. Practical application: offers diversified exposure with balanced risk contribution. Challenges: reliance on volatility estimates, sensitivity to correlation changes.
The ability of a system or algorithm to maintain performance as data volume or c… #
Example: distributing gradient‑boosted tree training across a Spark cluster for millions of tick data points. Practical application: enables real‑time analytics on large market datasets. Challenges: communication overhead, algorithmic bottlenecks.
The computational extraction of subjective information (positive, negative, neut… #
Example: scoring news headlines for bullish or bearish tone and feeding scores into a trading signal. Practical application: captures market mood for short‑term strategies. Challenges: sarcasm detection, domain‑specific lexicons, data latency.
A cooperative‑game concept that attributes a fair contribution to each feature i… #
Example: using SHAP to explain why a model assigned high default probability to a loan applicant. Practical application: enhances transparency for regulatory reporting. Challenges: computationally intensive for large models, approximation errors.
A supervised learning algorithm that finds the hyperplane separating classes wit… #
Example: classifying market regimes as “high‑volatility” or “low‑volatility”. Practical application: effective for small‑to‑medium sized, high‑dimensional financial data. Challenges: tuning kernel parameters, scaling to massive datasets.
A validation method that respects temporal ordering by training on past data and… #
Example: a 12‑month rolling window to evaluate a volatility‑forecasting model. Practical application: provides realistic performance estimates for sequential data. Challenges: limited training data in early windows, computational cost for many folds.
The process of breaking raw text into smaller units (tokens) such as words, sub‑… #
Example: tokenizing an earnings transcript before feeding it to a transformer model. Practical application: prepares unstructured data for AI analysis. Challenges: handling out‑of‑vocabulary terms, preserving context.
A technique where knowledge gained from training on one task or dataset is reuse… #
Example: fine‑tuning a BERT model pretrained on general news to specialize in financial disclosures. Practical application: accelerates model development and reduces data requirements. Challenges: domain mismatch, catastrophic forgetting.
A class of algorithms that infer patterns from data without explicit target labe… #
Example: using autoencoders to detect anomalous trading activity. Practical application: discovers hidden structures, such as latent factors. Challenges: evaluation metrics are less straightforward, risk of discovering spurious patterns.
A metric that quantifies how much the variance of an estimated regression coeffi… #
Example: a VIF of 12 for a factor indicates high redundancy. Practical application: guides feature selection to improve model stability. Challenges: thresholds are heuristic, may not capture non‑linear dependencies.
The empirical observation that large price changes tend to be followed by large… #
Example: modeling daily S&P 500 returns with a GARCH(1,1) process. Practical application: improves risk forecasts and option pricing. Challenges: structural breaks, model misspecification.
A regression technique that assigns different weights to observations, often inv… #
Example: weighting observations by inverse of trade volume variance when estimating a price impact model. Practical application: yields unbiased parameter estimates under non‑constant variance. Challenges: determining appropriate weights, sensitivity to outliers.
Dense, low‑dimensional vectors that capture semantic relationships between words… #
Example: using pre‑trained financial word embeddings to represent terms in earnings call transcripts. Practical application: improves NLP model performance on domain‑specific text. Challenges: domain shift, need for large corpora.
A paradigm where a model can correctly make predictions for classes it has never… #
Example: classifying a new type of market event using textual descriptions only. Practical application: enables rapid adaptation to novel market phenomena. Challenges: reliance on high‑quality semantic descriptors, limited accuracy compared to supervised alternatives.