Berat Arac Research
Published December 15, 2025 | Version v1
Preprint Open

Specialization Over Scale: Decision Layer Architectures for Efficient AI Systems

Authors/Creators

  • 1. Independent Researcher

Description

This preprint introduces the Decision Layer Architecture (DLA), a system-level framework that prioritizes specialization over scale in AI systems. The work presents the core architectural principles of DLA and provides preliminary estimates on cost and carbon efficiency.

Files

dla_technical_paper.pdf

Files (376.9 kB)

Name Size Download all
md5:ea54d7cf0efd326d677e03f7430c1415
376.9 kB Preview Download

Additional details

References

  • Kaplan, J., McCandlish, S., Henighan, T., et al. (2020). "Scaling Laws for Neural Language Models." arXiv:2001.08361. https://arxiv.org/abs/2001.08361
  • Chen, Z., Chen, M., & Wang, Y. (2025). "Revisiting Scaling Laws for Language Models." ACL Anthology. https://aclanthology.org/2025.acl-long.1163.pdf
  • Mishra, V. (2025). "Sustainability in large language model supply chains." Scientific Reports, 15(1). https: //www.nature.com/articles/s41598-025-17937-8
  • Husom, E. J., Goknil, A., Astekin, M., et al. (2025). "Sustainable LLM Inference for Edge AI: Evaluating Quantized LLMs for Energy Efficiency, Output Accuracy, and Inference Latency." ACM Transactions on Embedded Computing Systems. https://dl.acm.org/doi/abs/10.1145/3767742
  • Shazeer, N., Mirhoseini, A., Maziarz, K., et al. (2017). "Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer." arXiv:1701.06538. https://arxiv.org/abs/1701.06538
  • Zhang, H., Liu, Z., & Wang, Y. (2025). "A Survey on Mixture of Experts: Advancements, Challenges, and Future Directions." TechRxiv. https://www.techrxiv.org/doi/full/10.36227/techrxiv.173835706. 69246194
  • Sze, V., Chen, Y., & Zhang, H. (2024). "Conditional computation in neural networks: Principles and research trends." International Journal of Artificial Intelligence. https://arxiv.org/abs/2403.07965
  • Maconochie, J. (2024). "Beyond Scale: Towards Biologically Inspired Modular Architectures for Adaptive AI." Preprint. https://jamesmaconochie.com/assets/papers/ beyond-scale-modular-ai-maconochie.pdf
  • Langford, M. A., et al. (2022). "A Modular and Composable Approach to Develop Trusted AI Systems." IEEE International Conference on Autonomic Computing and Self-Organizing Systems (ACSOS). https: //ieeexplore.ieee.org/document/9935024/
  • Wang, J., et al. (2023). "Early-Exit Deep Neural Network - A Comprehensive Survey." ACM Computing Surveys. https://dl.acm.org/doi/10.1145/3698767
  • Lannelongue, L., Grealey, J., & Mok, S. (2021). "Green Algorithms: Quantifying the Carbon Emissions of Computation." Advanced Science, 8(12), 2100707. https://www.green-algorithms.org
  • Lacoste, A., Luccioni, A., Schmidt, V., & Dandres, T. (2019). "Quantifying the Carbon Emissions of Machine Learning." ML CO2 Impact. https://mlco2.github.io/impact