Wikis

Govern the use of recommendation algorithms across various sectors

Apply to providers of generative AI services and users of these services

Regulations regarding handling deep synthesis data and technologie

Initiated by the the U.S. Equal Employment Opportunity Commission (EEOC)

The world's first comprehensive law on artificial intelligence (AI)

Managing risks and promoting trustworthy AI systems in organizations

Tackling deceptive or unfair advertising claims related to AI

Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence

First part of an approach paper by NITI Aayog

Establishing a comprehensive legal framework for India's digital economy

To modernize and improve the government's data collection and management procedures

India's first national strategy for AI

Cybersecurity framework to guide critical sectors

India became a member of the Global Partnership on Artificial Intelligence (GPAI)

Constrains the model's predictions during the generation process to the top-k most probable tokens at each step

Parameter often associated with models that generate probability distributions

Approach for fine-tuning large-scale, pre-trained models

Process of converting continuous infinite input values from a large set to discrete finite output values in a smaller set

Model output generated that is not grounded in the input data but instead is generated imaginatively or erroneously

Capable of understanding and generating content across multiple data types or ‘modalities’

Learning technique used in generative AI to enhance the capabilities of models

Innovative state-space model (SSM) designed to handle complex, data-intensive sequences

The choice between open-source and closed-source models has emerged as a key debate within generative AI

Smallest unit or chunks of text that a model processes

Models designed to generate realistic, high-resolution images of varying quality.

Foundational blocks for various natural language processing (NLP) and generative AI tasks

Measures taken to ensure that AI is deployed in ways that do not harm humanity

Category of databases which store data as numerical representations of various complex forms of data

Machine learning models that utilize deep learning algorithms for understanding and analyzing natural language

Serve as a basis for a wide range of natural language processing (NLP), computer vision tasks, etc.

Generating synthetic data that imitate real-world data

Generate new data samples that resemble the patterns or distribution of the input data

Z-test is a statistically significant test used to determine whether two population means are different from each other.

The chi-square test is a statistical hypothesis test used to determine if there is a significant discrepancy between the expected and observed frequencies in one or more categories. 

The Kolmogorov-Smirnov statistic determines the probability of two samples coming from the same distribution. It does not make any assumptions about the underlying distribution of the data, making it a nonparametric test. 

The Wasserstein distance is a distance function for measuring the distance between two probability distributions.

The Jensen-Shannon distance measures the similarity between two probability distributions.

PSI is a measure of the degree to which the distribution of a population has shifted over time or between two different samples of a population in a single number.

KL divergence, also referred to as relative entropy, calculates the difference between two probability distributions.

ML audit ensures identifying associated risks and developing safeguard controls to avoid risks.

ML bias is a phenomenon where some aspects of datasets with equal significance are given more weight or representation than others, leading to skewed outcomes.

A practice of proactively examining and assessing crucial business data against quality control guidelines to maintain a high, consistent standard of data quality proactively

ML observability entails tracking the performance of ML system across its lifecycle, right from when it is being built, to pre and post-production.

MLOps is the discipline of deploying and maintaining machine learning models in production efficiently and reliably.

A statistical technique for analyzing the correlation between independent features or variables and a dependent feature, result, or variable.

Classification algorithm is a type of supervised learning that uses machine learning algorithms to learn how to categorize instances based on training data.

Model confidence score expresses how confident the machine learning model or NLP service is that the relevant intent has been accurately assigned

Thresholding is a technique for getting the most value out of a machine learning classifier.

An interpretable model that is trained to approximate the predictions of a black box model.

Enables teams to run hypothetical scenarios to understand what features need to be tweaked to obtain desired predictions

Similar cases help machine learning models discover groups of similar examples within the data.

Feature importance relates to methods for scoring each input feature for a particular model; the scores indicate the importance of each feature.

The LOCO method seeks to understand the significance of a particular feature for a model's prediction performance.

SHAP (SHapley Additive exPlanations) is a game theoretic approach for explaining the output of any machine learning model.

LIME approach attempts to comprehend the model by varying the input of data samples and observing how the predictions change.

AI explainability (XAI) refers to the process of explaining to an individual the decision-making process of an AI model

Bias is the situation where the model consistently predicts distorted results because of incorrect assumptions.

ML monitoring encompasses establishing alerts on key model performance metrics such as accuracy and drift.

Machine learning model accuracy is the measurement used to discover which model is best at recognising correlations and patterns between variables in a dataset.

Data processing involves transforming raw data into something meaningful that can be used by a machine learning model.

Precision is defined as the percentage of positive or relevant outcomes.

The percentage of actual positives that were accurately identified is known as Recall.

The percentage of true negatives the model correctly identifies is known as specificity.

The precision-recall curve depicts the tradeoff between precision and recalls for various thresholds.

The performance of the classification models for a given set of test data is evaluated using a confusion matrix.

The harmonic mean of precision and recall is used to calculate the F-measure, giving each the same weight.

ROC curve is a plot summarizing how well a binary classification model performs for the positive class.

Data drift is the change in model input data that causes a decline in model performance.

Model drift transpires when the ML models predictive ability "drifts" from the performance over the training period.