Researchers propose THPL framework for rainbow trout feeding in RAS using LLMs
Researchers introduced THPL, a generative feeding decision framework designed for rainbow trout in Recirculating Aquaculture Systems (RAS). The framework uses Fishsort to extract fish movement trajectories and compute an Activity Coefficient (AC) that measures feeding intensity. A Hierarchical Behavior Encoder (HBE) processes these trajectories into dual-evidence tokens representing physical and…
Key points
- THPL framework uses Fishsort and HBE to turn trout trajectories into dual-evidence tokens for LLM feeding decisions
- Decision accuracy improved from 33.33% to 96.67% with counterfactual mDPO over text-only baseline
- AC showed Spearman ρ = 0.925 (p < 0.001) correlation with expert-annotated feeding intensity
These tokens are combined with environmental data and expert rules to fine-tune a large language model via LoRA, then refined using counterfactual multimodal Direct Preference Optimization (mDPO) to improve causal reasoning. In tests, AC showed a strong correlation with expert-annotated feeding intensity (Spearman ρ = 0.925, p < 0.001). Decision accuracy rose from 33.33% with a text-only baseline to 93.33% with dual-evidence tokens, and further to 96.67% with counterfactual mDPO.
The framework also improved language quality metrics: METEOR increased from 58.10% to 85.30%, Self-BLEU-2 decreased from 58.79% to 52.88%, and Distinct-3 rose from 6.68% to 7.81%, indicating reduced templating bias and better output diversity. The researchers say THPL offers a new decision support paradigm for precision aquaculture by linking real-time fish behavior with LLM-based reasoning.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Google VP Yossi Matias says AI’s biggest impact may come from intersecting fields · 1 src
- Researchers adapt speech language model for simultaneous translation using prefix supervision · 1 src
- Study finds inductive prompting most consistent for LLM generalization in temporal extraction · 1 src
- AraBERT-based framework reaches 96.88% accuracy on Arabic DP ambiguity · 1 src
- Researchers introduce APDMem hierarchical memory for long-context LLM assistants · 1 src
Comments
via GitHub Discussions