Terminology

Multimodal Financial Foundation Models (MFFMs) is an intersection field of foundation models and finance. To facilitate readers from various backgrounds, we provide two lists of terminologies in below tables.

Table 1 Terminology for LLMs and agents

Key Terms	Explanations
Transformer	A transformer is a neural network architecture that utilizes the multi-head attention mechanism.
Large Language Model (LLM)	An LLM is a type of machine-learning model for human-like text understanding and generation.
Pre-training	The initial training phase where a model learns general features from a large dataset.
Fine-tuning	A subsequent training stage that adapts a pre-trained LLM to specific use cases, saving time and GPU hours.
Generative Pre-trained Transformer (GPT)	A family of LLMs based on the transformer architecture.
Prompt engineering	Crafting instructions to elicit the best possible output from an LLM.
Zero-Shot Prompting	Giving an LLM a task with no examples, relying on its prior knowledge.
Few-Shot Prompting	Guiding a generative model with a handful of examples for the task.
Chain-of-Thoughts (CoT)	A prompt strategy that directs a model to show intermediate reasoning steps.
In-Context Learning (ICL)	A paradigm where a model sees a few examples in the prompt and directly predicts the answer for new inputs.
Foundation Model	A large model trained on vast data that can be adapted across many downstream tasks.
FinLLM	A foundation model specialized for financial applications.
Multimodal	Involving multiple data modalities, such as text, image, audio, or video.
Retrieval-Augmented Generation (RAG)	Enhancing LLM outputs by retrieving information from external knowledge bases before generation.
Low-Rank Adaptation (LoRA)	A parameter-efficient fine-tuning technique that greatly reduces trainable parameters.
QLoRA	An extension of LoRA that quantizes model weights to 4-bit precision for memory efficiency.
Agent	An LLM-powered decision maker that tackles complex tasks using the model as its central engine.
Model Context Protocol (MCP)	A standard for connecting AI assistants to data repositories, business tools, and development environments.
Agent2Agent (A2A) Protocol	A standard for communication between independent AI agents.
Openwashing	Presenting models as open source when they are released under non-permissive licenses.

Table 2 Terminology for finance

Key Terms	Explanations
Earnings Conference Calls (ECCs)	Calls where a public company and stakeholders discuss the company’s financial results.
Monetary Policy Calls (MPCs)	Meetings in which central banks decide on monetary policy actions.
Environmental, Social, Governance (ESG)	An investment philosophy prioritizing environmental, social, and corporate-governance factors.
Financial Decision Making	Evaluating options, choosing among them, and taking actions (e.g., trading) on financial matters.
eXtensible Business Reporting Language (XBRL)	The global standard that powers digital business reporting.
Common Domain Model (CDM)	A standardized, machine-readable data and process model for the full lifecycle of financial products.
Robo-Advisor	An automated advisor providing algorithm-driven investment management without human intervention.
Digital Regulatory Reporting (DRR)	A cross-industry initiative aimed at transforming regulatory reporting infrastructure.
Greenwashing	Marketing that promotes superficial or misleading climate solutions, delaying credible action.