What’s New – BayesiaLab 11

BayesiaLab 11.6

Version 11.6 of BayesiaLab is the latest iteration of our flagship product that has been under continuous development for nearly 25 years. No other organization has invested as many resources in developing technologies around the Bayesian network paradigm.

Release 11.6 once again features many innovations, including the native integration of a LLM-based subject matter assistant (OpenAI, OpenAI GPT Assistants, Azure, Mistral, Anthropic, Google AI, Llama API, DeepSeek, Voyage AI).

This page provides an overview of all the new features introduced in BayesiaLab 11, covering versions 11.0 through 11.6. Only the features specific to version 11.6 are marked with the icon .

Download BayesiaLab 11.6

Hellixia

Hellixia Menu

Hellixia is the name of BayesiaLab’s subject matter assistant based on Generative AI. It provides a comprehensive suite of AI-driven tools designed to help you characterize and analyze a given problem domain. From causal modeling and semantic analysis to knowledge extraction, clustering, and document processing, Hellixia offers a structured approach to uncovering insights, identifying relationships, and enhancing decision-making.

Hellixia Features

Automatic Causal Network Generator

• The Automatic Causal Network Generator creates a Causal Bayesian Network from your question by tapping into the knowledge available in LLMs.
• A Risk-Centric Causal Network option can be used to generate a causal network specifically focused on characterizing the risks associated with the criterion, including triggering events, control actions, risk consequences, and mitigation actions.
• In this context, Automatic Causal Network Generator creates nodes, adds explanatory comments for each causal link, retrieves beliefs about causal effects (with values ranging from -100 to 100), and constructs the conditional probability tables.
• The resulting Causal Bayesian network is formatted automatically for easy interpretation.

Causal Network on Car Accidents

Risk-Centric Causal Network on Car Accidents

Automatic Semantic Network Generator

The Automatic Semantic Network Generator automatically extracts the main dimensions related to the user’s question using a set of selected keywords. It then creates a dataset with the corresponding embeddings and applies machine learning to generate a network that reflects the semantic proximities between the dimensions.

Semantic Network on Car Accidents

Document Analysis

Knowledge File and General Context

A Knowledge File is a file added as a detailed context for Large Language Models (LLMs), complementing the General Context to enhance domain-specific understanding.

Various formats are supported, including txt, pdf, Word, Excel, RTF, HTML, XML, PowerPoint, mp3, pcm, wav, aac, opus, flac.

A Preview feature is also available to verify that the file has been correctly interpreted before processing.

Preview of the Knowledge File

Six tools are available to perform a structured semantic analysis of your Knowledge Files.

• Semantic Flowchart Generator: Generates a graph where the key semantic concepts extracted from the knowledge file are connected to represent their sequential relationships.
• Causal Semantic Diagram Generator: Generates a graph where the key semantic concepts extracted from the knowledge file are connected to represent their causal relationships.
• Knowledge Graph Generator: Generates a graph where the key entities extracted from the knowledge file are connected to illustrate their semantic relationships, providing a meaningful framework for understanding associations between concepts.
• Causal Network Generator: Generates a Causal Network or Risk-Centric Causal Network from the knowledge file, using all the information contained in the file.
• Semantic Network Generator: Machine learns a Semantic Network based on the keywords selected by the user to capture the main dimensions of the knowledge file.
• Doc-to-Node: Creates a node for each selected knowledge file, where the node name is the filename and the node comment is the content of the file. It can also generate a data set with the corresponding embeddings and apply machine learning to create a network that reflects the semantic proximities between the node names or comments.

Doc-To-Node

Semantic Flowchart Generator

Generates a graph where the key semantic concepts extracted from the text, which is composed of the selected node name, long name, comment, and/or knowledge file, are connected to represent their sequential relationships.

Semantic Flowchart of Key Concepts from a Scientific Article on Actinic Lentigos

Causal Semantic Diagram Generator

Generates a graph where the key semantic concepts extracted from the text—composed of the selected node name, long name, comment, and/or knowledge file—are connected to represent their causal relationships.

Causal Semantic Diagram of Key Concepts from a Scientific Article on Actinic Lentigos

Knowledge Graph Generator

Generates a graph where the key entities extracted from the knowledge file are connected to illustrate their semantic relationships, providing a meaningful framework for understanding associations between concepts.

Knowledge Graph of Key Entities from a Scientific Article on Actinic Lentigos

Entity Relationship Finder

Automatically identifies potential relationships between selected nodes within a knowledge graph to discover meaningful connections between them.

Verbalize Relationships

Generates natural language descriptions of relationships by analyzing selected arcs in the graph, translating these connections into clear, human-readable explanations.

Causal Network Generator

This tool generates a Causal Bayesian Network or a Risk-Centric Causal Network focused on the chosen node. It generates new nodes, provides detailed comments for each causal link explaining the mechanism, determines causal effects (with values ranging from -100 to 100), constructs conditional probability tables, and can estimate prior probabilities for root nodes.

Causal Relationship Finder

Similar to the Causal Network Generator, this tool is designed to generate a Causal Network using a predefined set of nodes instead of centering around a single node and generating new nodes. It supports both supervised searches—focusing on a specific target node—and unsupervised searches, where potential causal connections are discovered across the predefined nodes.

Multi-Engine Causal Relationship Finder

The Multi-Engine Causal Relationship Finder is an extension of the Causal Relationship Finder, enhancing its capabilities by leveraging multiple Large Language Models (LLMs) for improved accuracy and robustness. Like the original version, it builds a Causal Network using a predefined set of nodes rather than centering around a single node and generating new ones. It also supports both supervised searches—focusing on a specific target node—and unsupervised searches, where potential causal connections are discovered across the predefined nodes. The use of multiple LLMs allows for a more diverse and comprehensive identification of causal relationships.

Causal Structural Priors

This tool evaluates whether the relationships represented by the selected arcs are causal. When causal relationships are identified, it generates Structural Priors, with values reflecting the confidence level in the existence of the relationship, along with comments describing the causal mechanisms. If the arcs follow the causal direction, the comments are associated with the arcs.

Pairwise Causal Link

This function evaluates whether a causal relationship exists between the two selected nodes and adds an arc if one is found. It quantifies the causal effect (ranging from -100 to 100), creates or updates the conditional probability table, and adds a comment to the arc explaining the causal mechanism. Additionally, it can generate the corresponding Structural Prior.

Root Priors Elicitor

Estimates prior probabilities of root nodes in networks with a target node using multiple LLMs. Each model acts as an independent expert, providing probability estimates with confidence levels and justifications. The system then aggregates these assessments to generate consensus probabilities, similar to human expert elicitation via BEKEE.

ICI Local Effects Elicitor

Applies the Multi-Engine approach to estimate an independent local effect, producing a continuous value instead of a probability distribution. Similar to the Root Priors Elicitor, this method mimics a brainstorming session, where each Large Language Model (LLM) acts as an independent expert, providing its own local effect estimation, a confidence level, and a justification of its assessment. The Target node serves as the reference point for these estimations, meaning the elicited local effects represent their influence on this specific target. The outputs from the LLMs are treated as expert assessments, comparable to those obtained from human experts via BEKEE. The system automatically computes a consensus, aggregating all estimated values and their associated confidence levels for a well-informed local effect estimation.

Dimension Elicitor

Identify key dimensions of a problem domain using a comprehensive set of keywords. For each identified dimension, the system creates a corresponding node with a detailed description.

There are now 165 keywords available, along with predefined groups of keywords tailored for specific use cases, such as capturing the useful dimensions of a Domain, conducting a Risk Analysis study, and facilitating the analysis of Scientific Articles, among others. These enhancements streamline the identification process and ensure more relevant and structured knowledge extraction.

Embedding Generator

This tool generates embeddings that capture the semantics of nodes using high-dimensional vectors. These vectors are stored in the data set associated with the graph, enabling the machine learning of semantic networks. If a data set already exists, the embeddings can either complete, replace, or be appended to it.

Keyword-Based Node Comment Generator

Enhance node content by leveraging a comprehensive keyword set to identify relevant dimensions within the problem domain. These dimensions are then added as comments to the corresponding nodes.

Definition-Based Comment Generator

Automatically creates descriptive node comments explaining the concepts each node represents within the network.

Condense/Elaborate Comments

Condenses or elaborates the comments associated with the selected nodes/arcs.

Elaborate Comment

Long Name Generator from Node Comment

Automatically generates long names that summarize the comments associated with the selected nodes.

Latent Variable State Labeler

Generates descriptive long names for the states of latent variables created during Data Clustering and Multiple Clustering. By analyzing the node’s characteristics and leveraging the Weight of Evidence, this feature assigns meaningful names to the variable states, enhancing clarity and interpretability in data-driven models.

Automatic Labeling of Clusters

Class Description Generator

Creates a class with the selected nodes and generates a descriptive summary of the corresponding factor based on the semantics of the associated nodes.

Semantic Variable Clustering

Create clusters (Factors) of nodes based on their semantic.

Image Generator

Produces icons that visually represent the information linked to the nodes.

Translator

The Translator converts various network elements—including names of nodes, states, and comments on nodes and arcs—into the chosen language.

Now, it can also leverage a specified Context and a Knowledge File to enhance translations, ensuring more accurate and domain-specific results. This improvement allows for better adaptation to specialized terminologies and nuanced meanings within a given problem domain.

Report Analyzer

Processes the output from the Relationship Analysis Report, including arc and node forces, and generates an HTML report that outlines the key dynamics of the domain represented by the network.

Report Analyzer

Speech-to-Text

Press the mic icon associated with any of BayesiaLab’s text fields and use the Speech-to-Text feature to enter your text (node names, comments, …).

Additionally, the mic icon in the main toolbar allows you to search BayesiaLab’s functions using voice commands, making it easier to quickly find and access the features you need.

Independence of Causal Influence

The Independence of Causal Influence (ICI) tool has been enhanced with several updates:

New Combination Functions

• SumPos(): An asymmetrical variation of the Sum function focusing on positive local mechanical effects.
• SumNeg(): A counterpart that emphasizes negative local mechanical effects.
• MinMax(): A function that implements the min method for negative values and the max method for positive ones.

ICI Wizard Enhancement

• A Condensed Display option has been introduced. This feature creates a network where the local effects are snapped to their parent and the combination nodes to their respective children.

Expert Knowledge Elicitation-Related Features:

• The Expert Editor has been renamed as the SMEs & BEKEE Session Manager.

• Subject Matter Experts (SMEs) can now be identified with specific colors for better differentiation.

• There’s an option to to send out invitation emails to the SMEs.

• In terms of qualitative knowledge elicitation, specifically the qualitative segment of the Delphi Method, you can now utilize Generate Notes from Comments in the Assessment Editor to produce Notes directly on the Graph Panel, derived from expert comments collected in BEKEE.

  Additionally, a Note Report feature is now available, allowing access to all note contents in an HTML report (Reports | Notes).

Expert Notes Generated from BEKEE Comments in BayesiaLab

• Generate Nodes from Comments, also available in the Assessment Editor, utilizes Hellixia for analyzing expert comments collected in BEKEE. This function automatically generates a Semantic Network by identifying key dimensions mentioned in the comments. When experts provide structured lists of dimensions (e.g., using bullet points), this feature efficiently extracts and organizes them into nodes. It significantly reduces the time required for analyzing notes, creating nodes, ensuring no duplicates or equivalent concepts exist, and even generates useful descriptive comments for defining each dimension—an essential step in well-organized brainstorming sessions that is often overlooked due to time constraints. Furthermore, each expert can use their own language, as the system can process multilingual inputs, ensuring inclusivity and accessibility in collaborative knowledge elicitation.

New BEKEE Elicitation Wizard

• Send Priors: When eliciting the probability distribution of a node, Send Priors allows dispatching its current distribution as a prior to all experts in BEKEE, serving as an alternative to the default uniform distribution.

  If the distribution being sent is a consensual distribution—derived from multiple expert-elicited distributions—Hellixia is used to generate an informative summary of all comments associated with these distributions. This summary is sent along with the consensual distribution to BEKEE, ensuring that experts not only see the agreed-upon prior but also a synthesis of the justifications provided by those who have already contributed their assessments. This workflow is the one used during our new brainstorming sessions, where we leverage multiple LLM engines as virtual experts to estimate local effects and provide justifications. The consensus derived from these LLMs is then sent to BEKEE, allowing human experts to review both the aggregated distribution and a summary of all LLM-generated justifications, fostering a more informed and collaborative decision-making process.

• The Assessment Mode setting allows you to choose the type of assessment that will be used in BEKEE, providing flexibility in expert elicitation. Experts can now provide their input using one of the following modes:

  • Probability Distribution Mode (Default) – Experts define a full probability distribution for the node.
  • Mean Value Mode (New) – Experts provide a numerical value, which is then transformed into a probability distribution using the Binary Mean Method. This approach is particularly useful when eliciting independent local effects, as it allows experts to express their assessments in a more intuitive way.
  • Both Modes – Experts can choose between defining a probability distribution or providing a numerical mean value that will be converted accordingly.

• Node Contextual Menu:

  • Generate from Assessments: this function facilitates the creation of distributions based on the weighted votes of chosen experts.
  • Generate Assessments: This feature uses the node’s current probability distribution to create an assessment associated with a selected expert. If Prior Weights are linked to the node, there’s an option to use these weights to determine the expert’s confidence level in the assessments.
  • Delete Zero-Confidence Assessments: this option removes all assessments in which the expert’s confidence level is set to 0.
  • Delete Assessments: this feature deletes the assessments linked to the selected experts.

• Hellinger Distance: Measures the distance between experts’ votes and a reference expert (usually the consensus).

• 2D/3D Mapping incorporates new metrics derived from experts’ assessments.

Formulas

The Formulas tab in the Node Editor now supports local variables.

Additionally, new functions have been introduced, such as:

• TriangularMD(v1, x), i.e., triangular membership degree in fuzzy logic (under Special Functions)

• Deciban(x): The deciban is a logarithmic unit — much like the decibel or the Richter scale — introduced by Alan Turing for expressing probabilities. It is a tenth of a ban, which is also known as the base-10 log odds (under Arithmetic Functions)

• Hellinger(v1, v2): The Hellinger distance is a measure of the similarity between two probability distributions (under Inference Functions)

• NoisySum(s, leak, v1, w1, vn, wn): Used for representing situations where the variable s is the weighted (wi) sum of its parents (vi) plus an additional noise term (leak) to model uncertainty or random fluctuations

• SwitchNoisyOr(s, leak, c1, p1, cn, pn): This function implements a modified Noisy-Or model that operates based on the combined effect of all pi values. The parameters ci represent conditions or boolean variables, while pi are their associated effects (positive or negative).

  • When the aggregated sum of pi values is positive, the function executes a Noisy-Or with an overall effect equal to this sum, effectively determining the probability of the True state.
  • Conversely, when the sum is negative, the function applies the Noisy-Or logic to the False state, adjusting the likelihood of the outcome being False according to this negative sum.

• DualNoisyOr(s, leak, c1, p1, cn, pn): This function implements a variant of the Noisy-Or model where the total effect is determined by combining individual effects (pi) from boolean conditions (ci). Each pi can be positive or negative, influencing the overall outcome.

  The function applies an AND operation between two components:

  • A Noisy-Or applied to the nodes that have a positive effect.
  • The negation of a Noisy-Or applied to the nodes that have a negative effect.

  This structure refines the probability computation by integrating both reinforcing and inhibiting influences.

• SingleMode(v): A function designed to ascertain whether the distribution of variable v is unimodal (under Inference Functions).

Weight of Evidence

Weight of Evidence now features four new types of analyses:

• Most/Least Relevant Explanations
• Most/Least Confirmatory Clues

Structural Learning Algorithms

The EQ-based learning algorithms are now disabled in situations where the score of an arc is not the same in both directions. This can occur due to Filtered States, Constraints, Structural Priors, etc. The assumption of equivalence is no longer theoretically valid in such contexts and could result in invalid networks including cycles.

Evidence Scenario Files

• The dataset associated with a network can now be exported into an Evidence Scenario file.
• Scenarios are now editable, allowing adjustments to the index, weight, and comments.
• A new Evidence Scenario Report is now available, offering a detailed description of the scenarios’ content.

Target Evaluation

The redesigned Target Evaluation function now features dedicated tabs for:

• Classification
• Posterior Probabilities
• Regression
• Triage

Graph Layout, Rendering and Editing

Dynamic Grid Layout

This innovative layout algorithm, particularly suitable for creating readable graphs featuring badges with associated comments, is also very practical for handling graphs created with Hellixia.

View Menu

Four new functions have been introduced to optimize the display of graphs. Users can now shrink or stretch graphs both vertically and horizontally, offering more flexibility in displaying the graph.

Position Menu

This new item has been introduced to enable the adjustment of the graphical layers of Nodes and Notes. It’s available via their Context Menus.

Horizontal and Vertical Stacking

These new alignment tools enable the positioning of the selected nodes horizontally or vertically, aligning them with minimal space between them.

Highlight a Class

From the Context Menu of a Note, you can select a Highlight a Class and the Note will automatically adjust its size and position to include all nodes in the selected class.

Arc Editor

By double-clicking an arc, you can edit the Comment associated with the arc as well as its rendering properties.

Edit Arc

New properties added:

• Show Comment Colors: An option that allows a color to be associated with the comment itself, rather than using the arc’s color.
• Comment Color Linked to Nodes: Allows associating two colors with the comment— the top color reflects the parent node, and the bottom color corresponds to the child node.

These enhancsements improve the visual organization and clarity of arc annotations.

Arc Comments

The width, layer, and position of Arc Comments can now be modified. Additionally, an Automatic Comment Layout algorithm (View Menu) is available for positioning comments.

Arc Comments with Comment Color Linked to Nodes

Color Link

This new feature, added to the Rendering Properties of Badges, Monitors, Bars, and Gauges, automatically applies a node’s associated color to the Name Background Color. Additionally, it automatically selects white for the Name Color on dark backgrounds and black on lighter ones.

Selection Zone

By pressing Z, you can create a selection zone, in which objects are always selected regardless of their current state, i.e., selected or not-selected.

Numerical Evidence Entry for Gauges and Bars

A new approach is introduced for entering numerical evidence through shift+click on a node. Utilize the M and B icons to select the Distribution Estimation Method (MinXEnt and Binary, respectively), with the three icon colors representing the Observation Type: No Fixing, Fix Mean, and Fix Probabilities, respectively.

Pseudo Root-Nodes

If a node only has Function Nodes as parents, making it a root node of its subnetwork, and the parents of these Function Nodes have fixed observed values, then the distribution of these pseudo root-nodes is also automatically set to fixed.

Boolean Conversion

Featured in the Tools menu, this function enables the conversion of selected nodes into boolean nodes.

2D Mapping

2D Mapping function has been enhanced to incorporate an additional dimension for node analysis: Font Size, supplementing the existing Node Size and Color dimensions. This enables font sizes to be proportional to the selected metric.

Node Analysis

The Node Analysis section has been enriched with numerous additional metrics, further enhancing its analysis capability:

• Mutual Information with Target Node
• Mutual Information with Target State
• Bayes Factor
• Normalized Bayes Factor
• Kullback-Leibler Divergence
• Normalized Kullback-Leibler
• Total Effect on Target
• Standardized Total Effect on Target
• Direct Effect on Target
• Standardized Direct Effect on Target
• Number of Assessments
• Assessment Completion Rate
• Maximum Assessment Divergence
• Overall Assessment Divergence
• Missing Value Rate

Comments

Comments associated with nodes are now displayed as you hover over them.

Hide Text

The option Hide Text for Ignored Nodes conceals the names of nodes that are Not Observable.

WebSimulator

The Class Selector is now available in the WebSimulator Editor, allowing users to select which nodes to display in the list of Available Nodes.

Available Nodes

Additionally, the color of each node is now displayed, enabling quick identification of nodes. This enhancement is particularly useful for networks with a large number of nodes, improving usability and organization.

Network Tabs

The tabs associated with each network, displayed at the bottom of the main BayesiaLab window, can now be reorganized, allowing users to customize their workspace for improved efficiency and navigation.

Learn More

• Hellixia User Guide
• BayesiaLab User Guide