Add new labels

[Irratzo]

• [x] Add labels
• [x] Update labels

Category details:

Update labels:

• [x] Change incorrectly hyphenated labels, including new labels below.
  • Affected labels: All labels prefixed with multi and pre.
  • multifidelity, multimodal, pretrained
  • Reasoning: These words should not be hyphenated, unless the following cases apply. 1) when the root word is a proper noun or starts with a capital, 2) when the root word begins with the same letter as the last letter in the prefix, 3) when the prefix is added to a compound, 4) when the prefix is added to a number, 5) when it is otherwise necessary to avoid confusion.
  • References.
  • Government of Canada. Writing Tips Plus. hyphens: pre-, post-. URL.
  • Microsoft Corporation. Microsoft Writing Style Guide. multi-. URL.
  • Microsoft Corporation. Microsoft Writing Style Guide. Hyphens. URL

New labels:

• [x] #335
• [x] #279
• [x] multifidelity
  • possible URLs: https://en.wikipedia.org/wiki/Multifidelity_simulation
  • add label to existing projects: MEGNet, MatGL (Materials Graph Library)
• [x] multimodal
  • possible URLs: https://en.wikipedia.org/wiki/Multimodal_learning
  • alternatives. vision, computer-vision could be added as well as subset.
  • add label to existing projects: #325.
• [x] neural-operator and pinn
  • possible URLs: https://en.wikipedia.org/wiki/Neural_operators, https://en.wikipedia.org/wiki/Physics-informed_neural_networks
  • add label to existing projects: NA
  • alternatives piml for physics-informed machine learning as wider label, would also include PiNNs etc.
• [x] ai-agent
  • possible URLs: https://en.wikipedia.org/wiki/Large_language_model#Agency, https://en.wikipedia.org/wiki/Intelligent_agent
  • add label to existing projects: whitead/paper-qa, ur-whitelab/chemcrow-public, CFN-softbio/SciBot, maykcaldas/MAPI_LLM
• [x] proprietary
  • possible URLs: https://en.wikipedia.org/wiki/Proprietary_software, https://simple.wikipedia.org/wiki/Closed_source
  • add label to existing projects: GNoME, #314, #339
  • alternatives: closed-source. But not all closed-source software is proprietary and vice-versa.
  • reasoning: GNoME and #339 are labeled "proprietary" on Matbench Discovery, as well.
• [x] FAIR
  • possible URLs: https://www.go-fair.org/fair-principles/
  • add labels to existing projects: zincware/IPSuite
  • Side note. Projects by the organization "FAIR Chemistry", formerly known as the Open Catalyst Project (ocp), have no relation to the FAIR principles, as far as I know.

Additional context:

[Irratzo]

Here is a list of all types of "multi" machine learning types, for reference, from Claude AI Sonnet 3.5.

Here's a list of common "multi" types of machine learning:

Here's a list of common "multi" types of machine learning, along with brief explanations to clarify their meanings and differences:

1. Multitask learning: Involves training a single model to perform multiple related tasks simultaneously. The model shares some parameters across tasks while having task-specific parameters as well. This can improve generalization and efficiency.

2. Multilabel learning: Deals with problems where each instance can be associated with multiple labels or classes simultaneously. For example, an image might contain both a cat and a dog.

3. Multiclass learning: Involves classifying instances into one of three or more classes. This is an extension of binary classification to multiple mutually exclusive categories.

4. Multitarget learning: Similar to multilabel learning, but focuses on predicting multiple continuous or ordinal target variables simultaneously, rather than discrete labels.

5. Multimodal learning: Involves processing and learning from multiple types of input data or "modalities" (e.g., text, images, audio) simultaneously to make predictions or perform tasks.

6. Multifidelity learning: Uses data or models of varying levels of accuracy or cost to improve learning efficiency, particularly useful in optimization and surrogate modeling.

7. Multi-instance learning: Deals with problems where labels are associated with sets or bags of instances, rather than individual instances.

8. Multiview learning: Utilizes multiple representations or "views" of the same data to improve learning performance, often by exploiting complementary information.

9. Multiagent learning: Involves multiple autonomous agents learning to interact and cooperate to solve complex problems or optimize collective performance.

10. Multioutput learning: Similar to multitarget learning, but more general. It can include both continuous and discrete outputs, encompassing multilabel and multitarget scenarios.

11. Multisource learning: Integrates data from multiple sources or domains to improve model performance and generalization.

These approaches often overlap and can be combined in various ways depending on the specific problem and dataset at hand.

I think the most important of these, with respect to current atomistic ML literature, are multifidelity, multimodal and maybe multitask learning.