Fishgrs Work | !!better!!
"Fishr" is an ICML 2022 machine learning framework designed for out-of-distribution (OOD) generalization, focusing on regularizing the variance of gradients across different training domains. This approach improves model robustness by enforcing consistent sensitivity across domains to learn invariant features rather than domain-specific shortcuts. Read the full paper at proceedings.mlr.press.
B. Drifting FADs (dFADs)
- Structure: These float freely with ocean currents. Historically made of bamboo rafts with hanging "tail" lines, modern dFADs often include sophisticated satellite buoys.
- Use Case: Dominant in industrial-scale tuna fisheries (purse seining).
- Advantage: Fishers can track the location and biomass under the FAD remotely via satellite, allowing them to target only the most profitable schools.
6. Management and Future Trends
Regional Fisheries Management Organizations (RFMOs) are actively working to regulate "FAD work" to ensure sustainability. Key initiatives include: fishgrs work
- FAD Limits: Imposing limits on the number of FADs a vessel can deploy at one time.
- Biodegradable FADs: Encouraging the transition from plastic/netting materials to natural fibers (cotton, bamboo, abaca) to reduce pollution and ghost fishing.
- Non-Entangling Designs: Mandating designs that prevent sharks and turtles from becoming tangled in the trailing lines.
- FAD Closures: Establishing specific months or areas where fishing on FADs is prohibited to allow fish stocks to recover.
Introduction
- Background: Genomic tools have transformed fisheries science by enabling high-resolution insight into population connectivity, local adaptation, introgression, and effective population size.
- Motivation: Traditional catch and tag methods lack the resolution for cryptic population structure and adaptive loci detection; FishGRS integrates genomic sampling with spatial and environmental data to fill these gaps.
- Scope: Define FishGRS as an end-to-end framework covering field sampling design, DNA extraction and library preparation, sequencing strategies, reference assembly considerations, variant calling, downstream population genomic analyses, and application to management decisions.
10. Challenges and Limitations
- Reference genomes: many fish species lack high‑quality references; structural variation and repeats complicate mapping.
- Transferability of scores: GRS or genomic prediction models trained in one population or environment often perform poorly in another due to LD differences and genotype–environment interactions.
- Sample size constraints: wild populations may be small, limiting power.
- Complex trait architectures: traits influenced by rare variants, epistasis, or GxE reduce predictability.
- Cost vs. benefit: balancing genotyping resolution, phenotyping effort, and expected genetic gain.
