Archive · 29 Jul 2026
Smaller AI Models Sufficient for Brain-Language Studies
Researchers from IIIT-H found that smaller AI models with about 3 billion parameters can predict human brain activity almost as well as larger models with up to 14 billion parameters.
They used techniques like quantisation and pruning to make the models lighter without significantly affecting their ability to predict brain activity.
This finding challenges the idea that larger models are always better.
Smaller models require less computing power, making brain-language studies cheaper and faster.
However, smaller models may not perform as well on conventional language tasks involving grammar and discourse.
This research could be a game changer for brain-decoding workflows and developing brain-computer interfaces.
Smaller AI models with 3 billion parameters perform almost as well as larger models with 14 billion parameters in predicting brain activity.
Techniques like quantisation and pruning reduce model size without substantially affecting brain activity predictions.
Smaller models require less computing resources, making brain-language studies potentially cheaper and faster.
Smaller models may perform worse on conventional language tasks involving grammar, discourse, and morphology.
The research challenges the notion that larger models are always better for brain prediction accuracy.
- Who
- Prof. Bapi Raju and PhD researcher Vijay Rowtula from IIIT-H
- What
- Research on the effectiveness of smaller AI models in brain-language studies
- When
- Presented at the International Conference on Machine Learning (ICML) in Seoul
- Where
- International Institute of Information Technology Hyderabad (IIIT-H)
- Why
- To understand if smaller AI models can effectively predict human brain activity and reduce computational resources needed for brain research
Smaller AI Models
Larger AI Models
Model Performance
Smaller AI Models
Smaller models with about 3 billion parameters perform almost as well as larger models with up to 14 billion parameters in predicting brain activity.
Larger AI Models
Larger models were previously thought to improve brain prediction accuracy by about 15 percent.
Computational Resources
Smaller AI Models
Smaller models require less memory and computing resources, making brain-language studies potentially cheaper and faster.
Larger AI Models
Larger models require more computational power and resources.
Linguistic Competence
Smaller AI Models
Smaller models maintain brain alignment but may perform worse on conventional language tasks involving grammar, discourse, and morphology.
Larger AI Models
Larger models may perform better on conventional language tasks.
- Research Institution
- International Institute of Information Technology Hyderabad (IIIT-H)
- Researchers
- Prof. Bapi Raju, PhD researcher Vijay Rowtula
- Conference
- International Conference on Machine Learning (ICML) in Seoul
- Model Parameters
- 3 billion to 14 billion
- Key Techniques
- Quantisation, Pruning
- Potential Impact
- Cheaper and faster brain-language studies, game changer for brain-decoding workflows