pith. sign in

arxiv: 2604.07602 · v1 · submitted 2026-04-08 · 💻 cs.NE · cs.CE· q-bio.NC

The Principle of Maximum Heterogeneity Optimises Productivity in Distributed Production Systems Across Biology, Economics, and Computing

Guillhem Artis , Danyal Akarca , Jascha Achterberg This is my paper

Pith reviewed 2026-05-10 16:49 UTC · model grok-4.3

classification 💻 cs.NE cs.CEq-bio.NC
keywords distributed production systemsheterogeneityproductivity optimisationspecialisationcommunication topologynested systemsAI compute designcross-disciplinary unification
0
0 comments X

The pith

Any performance-optimising distributed production system converges toward greater agent heterogeneity within environmental limits and communication-determined scales.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proposes a single model of distributed production systems that unifies how heterogeneity, resource constraints, communication topology, and task structure shape productivity and robustness in economies, brains, ecosystems, and computers. It advances the Principle of Maximum Heterogeneity as the governing rule: systems that optimise performance develop increasing differences among agents, with environmental demands capping the useful degree of difference and topology controlling how widely those differences spread. The same rule is said to operate recursively inside nested subsystems. If accurate, the model supplies a practical guide for redesigning large computing clusters that run AI workloads.

Core claim

The central claim is the Principle of Maximum Heterogeneity: any distributed production system optimising for performance will converge on an increasingly heterogeneous configuration; environmental demands place an upper bound on the degree of heterogeneity required; and the communication topology determines the spatial scale over which heterogeneity spreads, with this principle applying recursively across all layers of nested production systems.

What carries the argument

The Principle of Maximum Heterogeneity, which states that performance optimisation drives increasing agent heterogeneity in distributed production systems, subject to environmental bounds and communication topology.

Load-bearing premise

That well-understood findings from biology, economics, neuroscience, and computing can be captured in one simple joint model of distributed production systems.

What would settle it

A controlled distributed system in which forcing greater homogeneity under fixed environmental demands and topology raises rather than lowers measured productivity.

Figures

Figures reproduced from arXiv: 2604.07602 by Danyal Akarca, Guillhem Artis, Jascha Achterberg.

Figure 1
Figure 1. Figure 1: Intuitive visualisation of the distributed production system model in which each [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Intuitive visualisation of the distributed production system model with con [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: According to the theory of adaptive radiations, as species proliferate in a new [PITH_FULL_IMAGE:figures/full_fig_p019_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Homogeneous and heterogeneous temporal biomass production across unimodal [PITH_FULL_IMAGE:figures/full_fig_p020_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Agent populations naturally adapt their filters to match local stimulus distribu [PITH_FULL_IMAGE:figures/full_fig_p022_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Network topology underlying the MD system experiment. [PITH_FULL_IMAGE:figures/full_fig_p023_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The model converges on brain-like structure-function topology. [PITH_FULL_IMAGE:figures/full_fig_p023_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Three-tier network topology. The core agent (green) connects to all penumbra agents (blue), each of which links to three peripheral agents (purple) that have no other connections. The model recovers the expected gradient: core agents converge on the broadest skill distribu￾tions, penumbra agents on intermediate ones, and peripheral agents on the narrowest ( [PITH_FULL_IMAGE:figures/full_fig_p024_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Specialisation increases from core to periphery. [PITH_FULL_IMAGE:figures/full_fig_p025_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: displays the optimised production profiles. Under autarchy (top row), each country converges toward a broad, generalist production capacity that attempts to cover both demand peaks simultaneously. The resulting individual profiles are nearly identical and relatively flat, reflecting the impossibility for a single agent to efficiently serve two separated demand modes. Under trade (bottom row), each country… view at source ↗
Figure 11
Figure 11. Figure 11: Trading between countries increase realised consumption. [PITH_FULL_IMAGE:figures/full_fig_p026_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Productivity is positively correlated with heterogeneity and specialisation. [PITH_FULL_IMAGE:figures/full_fig_p028_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Labour market efficiency determines the specialisation of hired co-workers. [PITH_FULL_IMAGE:figures/full_fig_p029_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Per-market-evolution mean return (left) and coefficient of variation (right) [PITH_FULL_IMAGE:figures/full_fig_p030_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Learning performance evolution with heterogeneity and resource. [PITH_FULL_IMAGE:figures/full_fig_p032_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Scaling behaviour of the distributed production model. [PITH_FULL_IMAGE:figures/full_fig_p033_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Performance and robustness matrices for the homogeneous and heterogeneous [PITH_FULL_IMAGE:figures/full_fig_p034_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Heterogeneity and specialisation scale with the system size. [PITH_FULL_IMAGE:figures/full_fig_p036_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Logistic scaling of heterogeneity with system size. [PITH_FULL_IMAGE:figures/full_fig_p036_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Heterogeneity and specialisation scale with the system size. [PITH_FULL_IMAGE:figures/full_fig_p037_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: The low-level heterogeneity of the workload has the highest correlation with [PITH_FULL_IMAGE:figures/full_fig_p038_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: The two strongest univariate predictors of system heterogeneity. [PITH_FULL_IMAGE:figures/full_fig_p038_22.png] view at source ↗
Figure 23
Figure 23. Figure 23: Effect of the maximum communication component size. [PITH_FULL_IMAGE:figures/full_fig_p040_23.png] view at source ↗
Figure 24
Figure 24. Figure 24: Optimised agent positions for each component size. [PITH_FULL_IMAGE:figures/full_fig_p040_24.png] view at source ↗
Figure 25
Figure 25. Figure 25: Agent positions coloured by specialisation level. [PITH_FULL_IMAGE:figures/full_fig_p041_25.png] view at source ↗
Figure 26
Figure 26. Figure 26: Locality of heterogeneity induced by communication costs. [PITH_FULL_IMAGE:figures/full_fig_p042_26.png] view at source ↗
Figure 27
Figure 27. Figure 27: The efficient system-level architecture is a heterogeneous architecture. [PITH_FULL_IMAGE:figures/full_fig_p043_27.png] view at source ↗
Figure 28
Figure 28. Figure 28: Performance and heterogeneity decrease concurrently with stiffness. [PITH_FULL_IMAGE:figures/full_fig_p044_28.png] view at source ↗
Figure 29
Figure 29. Figure 29: Summary robustness for the catalogue suited to the homogeneous system [PITH_FULL_IMAGE:figures/full_fig_p045_29.png] view at source ↗
Figure 30
Figure 30. Figure 30: Workload-production comparison across different hardware instantiations. [PITH_FULL_IMAGE:figures/full_fig_p051_30.png] view at source ↗
Figure 31
Figure 31. Figure 31: Summary of system performance across hardware configurations. [PITH_FULL_IMAGE:figures/full_fig_p052_31.png] view at source ↗
Figure 32
Figure 32. Figure 32: Heterogeneous hardware reaches peak productivity at smaller system sizes. [PITH_FULL_IMAGE:figures/full_fig_p053_32.png] view at source ↗
Figure 33
Figure 33. Figure 33: Heterogeneous hardware is more performant than homogeneous hardware with [PITH_FULL_IMAGE:figures/full_fig_p054_33.png] view at source ↗
Figure 34
Figure 34. Figure 34: Comprehensive overview of tasks used in the model: (a) general principles, (b) multi￾Gaussian scaling, and (c) robustness testing [PITH_FULL_IMAGE:figures/full_fig_p073_34.png] view at source ↗
Figure 35
Figure 35. Figure 35: Overview of workloads used for extracting the efficiency principle. D.2 Metrics to characterise workload during analyses of principles For analyses in section 4.2, we have studied the effect of different features of the task onto the optimised system. We detail them in this appendix subsection. Let W0(θ) be the task demand profile defined on the torus, discretised over a grid of L = 2048 points with spaci… view at source ↗
Figure 36
Figure 36. Figure 36: Comparison of specialisation levels under different agent weighting schemes. To observe the comparison we overlay the two preceding results to visualise the specialisation deviation as a function of the weight deviations: δS = S(α) − S( ¯α) = (S(α1) − S(¯α), . . . , S(αN ) − S( ¯α)) 2 1 0 1 2 Weights' deviations, i 1.5 2.0 2.5 3.0 Specialisation Mean-field Base -0.62 0 1.13 Specialisation deviation [PITH… view at source ↗
Figure 37
Figure 37. Figure 37: Specialisation deviations with deviations of agents’ weights. 77 [PITH_FULL_IMAGE:figures/full_fig_p077_37.png] view at source ↗
Figure 38
Figure 38. Figure 38: Example production time series for the unimodal catalogue (task 4) under wave, Brownian motion, and extreme-event regimes. Blue: homogeneous system; purple: heterogeneous system. 0.00 0.01 0.02 0.03 0.04 0.05 Mean production 0.0 0.2 0.4 0.6 0.8 1.0 1.2 Coefficient of variation Wave Brownian motion Extreme event Homogeneous (Light) Heterogeneous (Dark) 0.000 0.005 0.010 0.015 0.020 0.025 0.030 0.035 Minimu… view at source ↗
Figure 39
Figure 39. Figure 39: Summary robustness statistics (mean, CV, minimum) averaged across all 16 unimodal tasks. Hatched bars: homogeneous; solid bars: heterogeneous. Colours indicate regime (blue: wave; purple: BM; dark blue: extreme). Diverse catalogue. When the task catalogue includes structurally diverse demands—uniform densities, bimodal and multimodal configurations with varying separations and asymmetries—the heterogeneou… view at source ↗
Figure 40
Figure 40. Figure 40: Example production time series for the diverse catalogue (task 4) under wave, Brownian motion, and extreme-event regimes. 0.000 0.005 0.010 0.015 0.020 0.025 Mean production 0.0 0.1 0.2 0.3 0.4 0.5 Coefficient of variation Wave Brownian motion Extreme event Homogeneous (Light) Heterogeneous (Dark) 0.000 0.005 0.010 0.015 0.020 Minimum production [PITH_FULL_IMAGE:figures/full_fig_p080_40.png] view at source ↗
Figure 41
Figure 41. Figure 41: Summary robustness statistics for the diverse catalogue (16 tasks spanning uniform through quintmodal configurations). Homo-suited catalogue: robustness despite initial disadvantage. The most revealing experiment uses the homo-suited catalogue, where all tasks are initially centred at µ = π—the exact specialisation of the homogeneous system—so that the homogeneous system begins with a large production adv… view at source ↗
Figure 42
Figure 42. Figure 42: Example production time series for the homo-suited catalogue (task 4). The homogeneous system starts with high production (task initially at µ = π) but suffers large swings; the heterogeneous system is steadier. 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 Mean production 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 Coefficient of variation Wave Brownian motion Extreme event Homogeneous Heterogeneous 0.0000 0.0025 0.0050 0… view at source ↗
Figure 43
Figure 43. Figure 43: Summary robustness for the homo-suited catalogue. Despite lower mean production, the heterogeneous system exhibits significantly lower CV and higher minimum production under extreme events. 81 [PITH_FULL_IMAGE:figures/full_fig_p081_43.png] view at source ↗
read the original abstract

The world is full of systems of distributed agents, collaborating and competing in complex ways: firms and workers specialise within economies, neurons adapt their tuning across brain circuits, and species compete and coexist within ecosystems. In that context, individual research fields built theories explaining how comparative advantage drives trade specialisation, how balanced neural representations emerge from sensory coding, and how biodiversity sustains ecological productivity. Here we propose that many of these well-understood findings across fields can be captured in one simple joint cross-disciplinary model, which we call the Distributed Production System. It captures how agent heterogeneity, resource constraints, communication topology, and task structure jointly determine the productivity, efficiency, and robustness of distributed systems across biology, economics, neuroscience, and computing. This model reveals that a small set of underlying laws generates the complex dynamics observed across fields. These can be summarised in our Principle of Maximum Heterogeneity: any distributed production system optimising for performance will converge on an increasingly heterogeneous configuration; environmental demands place an upper bound on the degree of heterogeneity required; and the communication topology determines the spatial scale over which heterogeneity spreads, with this principle applying recursively across all layers of nested production systems. Beyond explaining existing systems, these principles act as a blueprint for constructing ideal ones. We demonstrate this by suggesting specific redesigns for compute systems executing large-scale AI. In total, The Principle of Maximum Heterogeneity reveals a unique convergence of complex phenomena across fields onto simple underlying design principles with important predictive value for future distributed production systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper proposes a unified 'Distributed Production System' model integrating agent heterogeneity, resource constraints, communication topology, and task structure to explain productivity, efficiency, and robustness across biology, economics, neuroscience, and computing. It introduces the Principle of Maximum Heterogeneity: performance-optimizing systems converge on increasingly heterogeneous configurations, with environmental demands bounding the required heterogeneity and communication topology setting the spatial scale, applying recursively to nested systems. The principle is positioned as explanatory for existing phenomena (e.g., specialization, neural tuning, biodiversity) and as a blueprint for redesigning systems such as large-scale AI compute.

Significance. If the model were formally specified and the principle derived with validation against domain-specific results, the work could offer a cross-disciplinary framework with predictive value for distributed systems and practical guidance for AI infrastructure. The synthesis attempt across fields is ambitious and could encourage interdisciplinary thinking if the unification is demonstrated rather than asserted.

major comments (3)
  1. [Abstract] Abstract: the manuscript states that the Distributed Production System 'captures how agent heterogeneity, resource constraints, communication topology, and task structure jointly determine' outcomes and 'reveals' the Principle of Maximum Heterogeneity, yet supplies no state variables, objective function, equations, or derivation steps showing how the principle follows from these elements. This is load-bearing for the central claim.
  2. [Abstract] Abstract: the unification claim that 'many of these well-understood findings across fields can be captured in one simple joint cross-disciplinary model' is asserted without reproducing or deriving any specific established result (e.g., comparative advantage, balanced neural representations, or biodiversity-productivity relations) inside the model.
  3. [Abstract] Abstract: the application to 'suggesting specific redesigns for compute systems executing large-scale AI' is mentioned but no concrete redesigns, metrics, or evaluation steps are provided to support the claim of improved productivity or robustness.
minor comments (2)
  1. [Abstract] The description of the Principle of Maximum Heterogeneity is presented in a single dense sentence; breaking it into enumerated components would improve readability.
  2. [Abstract] Terms such as 'nested production systems' and 'spatial scale over which heterogeneity spreads' are introduced without explicit definition or prior context.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. These highlight opportunities to make the formal structure, unification, and applications more explicit. We address each major comment below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the manuscript states that the Distributed Production System 'captures how agent heterogeneity, resource constraints, communication topology, and task structure jointly determine' outcomes and 'reveals' the Principle of Maximum Heterogeneity, yet supplies no state variables, objective function, equations, or derivation steps showing how the principle follows from these elements. This is load-bearing for the central claim.

    Authors: We agree that the abstract does not reference the formal elements. The full manuscript defines the Distributed Production System with state variables for agent heterogeneity distributions, resource constraint vectors, communication topology as adjacency matrices, and task structures as allocation functions. Productivity is the objective function to be maximized subject to these constraints and environmental bounds. The Principle of Maximum Heterogeneity is obtained by showing that the optimum occurs at the highest feasible heterogeneity level, with topology determining the spatial extent and recursion applying to nested subsystems. We will revise the abstract to include a concise statement of these components and reference the derivation steps in the main text. revision: yes

  2. Referee: [Abstract] Abstract: the unification claim that 'many of these well-understood findings across fields can be captured in one simple joint cross-disciplinary model' is asserted without reproducing or deriving any specific established result (e.g., comparative advantage, balanced neural representations, or biodiversity-productivity relations) inside the model.

    Authors: We accept that explicit reproduction of domain results is needed to substantiate the unification. In revision we will add a section that derives simplified instances inside the model: a two-agent case recovering comparative advantage via optimal heterogeneity under resource constraints; a sensory coding example yielding balanced neural representations as the heterogeneity optimum; and a resource-partitioning case reproducing positive biodiversity-productivity relations. These will be presented as direct consequences of the same optimization rather than assertions. revision: yes

  3. Referee: [Abstract] Abstract: the application to 'suggesting specific redesigns for compute systems executing large-scale AI' is mentioned but no concrete redesigns, metrics, or evaluation steps are provided to support the claim of improved productivity or robustness.

    Authors: The manuscript outlines redesigns such as heterogeneous processor pools and topology-aware interconnects that increase feasible heterogeneity. To make this concrete we will expand the relevant section with explicit metrics (productivity as aggregate throughput, robustness as performance under node failure) and evaluation steps consisting of agent-based simulations of the DPS model on representative large-scale training workloads, comparing against homogeneous baselines. revision: yes

Circularity Check

0 steps flagged

No load-bearing derivation chain present to inspect for circularity

full rationale

The provided abstract and context describe a proposed model and principle but supply no equations, state variables, objective function, or explicit derivation steps. Without a concrete chain of the form 'from inputs A we derive result B via equation X' it is impossible to exhibit any reduction of a claimed prediction to its inputs by construction. The paper therefore cannot be scored for circularity under the required criteria; it is treated as self-contained in the absence of inspectable steps.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 2 invented entities

The central claim rests on the unproven assertion that disparate field-specific findings reduce to one simple model; the Principle of Maximum Heterogeneity is introduced as a new entity without independent evidence or derivation.

axioms (1)
  • domain assumption Distributed production systems optimize for performance
    Invoked as the driver that causes convergence to maximum heterogeneity.
invented entities (2)
  • Principle of Maximum Heterogeneity no independent evidence
    purpose: To explain and predict optimization behavior across all distributed production systems
    Newly proposed law with no external falsifiable handle or derivation shown.
  • Distributed Production System no independent evidence
    purpose: Unified cross-disciplinary model capturing agent heterogeneity, resources, topology, and tasks
    Newly introduced modeling framework without shown reduction to prior equations.

pith-pipeline@v0.9.0 · 5586 in / 1417 out tokens · 44195 ms · 2026-05-10T16:49:20.662305+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

110 extracted references · 110 canonical work pages · 2 internal anchors

  1. [1]

    Princeton University Press, Princeton, NJ, 1944

    John von Neumann and Oskar Morgenstern.Theory of Games and Economic Behavior. Princeton University Press, Princeton, NJ, 1944

    work page 1944

  2. [2]

    Non-cooperative games.Annals of Mathematics, 54(2):286–295, 1951

    John Nash. Non-cooperative games.Annals of Mathematics, 54(2):286–295, 1951

    work page 1951

  3. [3]

    The winner’s curse and public information in common value auctions.The American economic review, pages 894–920, 1986

    John H Kagel and Dan Levin. The winner’s curse and public information in common value auctions.The American economic review, pages 894–920, 1986

    work page 1986

  4. [4]

    Image scoring and cooperation in a cleaner fish mutualism.Nature, 441(7096):975–978, 2006

    Redouan Bshary and Alexandra S Grutter. Image scoring and cooperation in a cleaner fish mutualism.Nature, 441(7096):975–978, 2006

    work page 2006

  5. [5]

    A test of the producer- scrounger foraging game in captive flocks of spice finches, loncbura punctulata.Behavioral Ecology and Sociobiology, 34(4):251–256, 1994

    Luc-Alain Giraldeau, Catherine Soos, and Guy Beauchamp. A test of the producer- scrounger foraging game in captive flocks of spice finches, loncbura punctulata.Behavioral Ecology and Sociobiology, 34(4):251–256, 1994

    work page 1994

  6. [6]

    Human cortical networks trade communication efficiency for computational reliability.bioRxiv, pages 2025–12, 2025

    Kayson Fakhar, Danyal Akarca, Andrea I Luppi, Stuart Oldham, Fatemeh Hadaeghi, Petra E Vértes, Ed Bullmore, Claus Hilgetag, and Duncan Astle. Human cortical networks trade communication efficiency for computational reliability.bioRxiv, pages 2025–12, 2025

    work page 2025

  7. [7]

    A mathematical theory of communication.The Bell System Technical Journal, 27(3):379–423, 1948

    Claude E Shannon. A mathematical theory of communication.The Bell System Technical Journal, 27(3):379–423, 1948

    work page 1948

  8. [8]

    A general model for the origin of allometric scaling laws in biology.Science, 276(5309):122–126, 1997

    Geoffrey B West, James H Brown, and Brian J Enquist. A general model for the origin of allometric scaling laws in biology.Science, 276(5309):122–126, 1997

    work page 1997

  9. [9]

    Global modules robustly emerge from local interactions and smooth gradients.Nature, 640(8057):155–164, 2025

    Mikail Khona, Sarthak Chandra, and Ila Fiete. Global modules robustly emerge from local interactions and smooth gradients.Nature, 640(8057):155–164, 2025

    work page 2025

  10. [10]

    Lukas B. Freund. Superstar teams. CESifo Working Paper 12303, Center for Economic Studies and Ifo Institute (CESifo), 1 2025

    work page 2025

  11. [11]

    Brain-like process- ing pathways form in models with heterogeneous experts

    Jack Cook, Danyal Akarca, Rui Ponte Costa, and Jascha Achterberg. Brain-like process- ing pathways form in models with heterogeneous experts. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025

    work page 2025

  12. [12]

    Putting ricardo to work.Journal of Economic Perspectives, 26(2):65–90, 2012

    Jonathan Eaton and Samuel Kortum. Putting ricardo to work.Journal of Economic Perspectives, 26(2):65–90, 2012

    work page 2012

  13. [13]

    How heterogeneity shapes dynamics and computation in the brain.Neuron, 2025

    David Dahmen, Axel Hutt, Giacomo Indiveri, Ann Kennedy, Jeremie Lefebvre, Luca Maz- zucato, Adilson E Motter, Rishikesh Narayanan, Melika Payvand, Henrike Planert, et al. How heterogeneity shapes dynamics and computation in the brain.Neuron, 2025

    work page 2025

  14. [14]

    Deep-learning-assisted simulation of a cortical circuit: integrating anatomy, physiology and function.bioRxiv, 2026

    Shinya Ito, Darrell Haufler, Javier Galván Fraile, Kael Dai, Joseph Aman, Guozhang Chen, Claudio Mirasso, Wolfgang Maass, and Anton Arkhipov. Deep-learning-assisted simulation of a cortical circuit: integrating anatomy, physiology and function.bioRxiv, 2026

    work page 2026

  15. [15]

    Spa- tially embedded recurrent neural networks reveal widespread links between structural and functional neuroscience findings.Nature Machine Intelligence, 5(12):1369–1381, 2023

    Jascha Achterberg, Danyal Akarca, DJ Strouse, John Duncan, and Duncan E Astle. Spa- tially embedded recurrent neural networks reveal widespread links between structural and functional neuroscience findings.Nature Machine Intelligence, 5(12):1369–1381, 2023

    work page 2023

  16. [16]

    Computational ratio- nality: A converging paradigm for intelligence in brains, minds, and machines.Science, 349(6245):273–278, 2015

    Samuel J Gershman, Eric J Horvitz, and Joshua B Tenenbaum. Computational ratio- nality: A converging paradigm for intelligence in brains, minds, and machines.Science, 349(6245):273–278, 2015. 82 The Principle of Maximum Heterogeneity Optimises Productivity in Distributed Production Systems

    work page 2015

  17. [17]

    Mental labour.Nature human behaviour, 2(12):899– 908, 2018

    Wouter Kool and Matthew Botvinick. Mental labour.Nature human behaviour, 2(12):899– 908, 2018

    work page 2018

  18. [18]

    arXiv preprint 22 arXiv:2510.27434 (2025)

    Pengfei Sun, Jascha Achterberg, Zhe Su, Dan FM Goodman, and Danyal Akarca. Ex- ploiting heterogeneous delays for efficient computation in low-bit neural networks.arXiv preprint arXiv:2510.27434, 2025

    work page arXiv 2025

  19. [19]

    Building artificial neural circuits for domain-general cognition: a primer on brain-inspired systems-level architecture

    JaschaAchterberg, DanyalAkarca, MoatazAssem, MoritzHeimbach, DuncanEAstle, and John Duncan. Building artificial neural circuits for domain-general cognition: a primer on brain-inspired systems-level architecture. InAAAI 2023 Spring Symposium on the Evaluation and Design of Generalist Systems (EDGeS), 2023

    work page 2023

  20. [20]

    A generative network model of neurodevelopmental diversity in structural brain organization.Nature communications, 12(1):4216, 2021

    Danyal Akarca, Petra E Vértes, Edward T Bullmore, and Duncan E Astle. A generative network model of neurodevelopmental diversity in structural brain organization.Nature communications, 12(1):4216, 2021

    work page 2021

  21. [21]

    Production networks: A primer.Annual Review of Economics, 11(1):635–663, 2019

    Vasco M Carvalho and Alireza Tahbaz-Salehi. Production networks: A primer.Annual Review of Economics, 11(1):635–663, 2019

    work page 2019

  22. [22]

    Depth-width tradeoffs in approximating natural functions with neural networks

    Itay Safran and Ohad Shamir. Depth-width tradeoffs in approximating natural functions with neural networks. InInternational conference on machine learning, pages 2979–2987. PMLR, 2017

    work page 2017

  23. [23]

    Neural modularity helps organ- isms evolve to learn new skills without forgetting old skills.PLOS Computational Biology, 11(4):e1004128, 2015

    Kai Olav Ellefsen, Jean-Baptiste Mouret, and Jeff Clune. Neural modularity helps organ- isms evolve to learn new skills without forgetting old skills.PLOS Computational Biology, 11(4):e1004128, 2015

    work page 2015

  24. [24]

    The impact of trade on organization and productivity.The quarterly journal of economics, 127(3):1393–1467, 2012

    Lorenzo Caliendo and Esteban Rossi-Hansberg. The impact of trade on organization and productivity.The quarterly journal of economics, 127(3):1393–1467, 2012

    work page 2012

  25. [25]

    The heterogeneity–diversity–system perfor- mance nexus.National Science Review, 10(7):nwad109, 2023

    Nico Eisenhauer, Raúl Ochoa-Hueso, et al. The heterogeneity–diversity–system perfor- mance nexus.National Science Review, 10(7):nwad109, 2023

    work page 2023

  26. [26]

    Heterogeneity and efficiency in the brain.Proceedings of the IEEE, 103(8):1346–1358, 2015

    Vijay Balasubramanian. Heterogeneity and efficiency in the brain.Proceedings of the IEEE, 103(8):1346–1358, 2015

    work page 2015

  27. [27]

    M. O. Hill. Diversity and evenness: A unifying notation and its consequences.Ecology, 54(2):427–432, 1973

    work page 1973

  28. [28]

    Entropy and diversity.Oikos, 113(2):363–375, 2006

    Lou Jost. Entropy and diversity.Oikos, 113(2):363–375, 2006

    work page 2006

  29. [29]

    Mismeasuring biological diversity: Response to hoffmann and hoffmann (2008)

    Lou Jost. Mismeasuring biological diversity: Response to hoffmann and hoffmann (2008). Ecological Economics, 68(4):925–928, 2009

    work page 2008

  30. [30]

    Tom Leinster and Christina A. Cobbold. Measuring diversity: The importance of species similarity.Ecology, 93(3):477–489, 2012. Concepts and Synthesis. Corresponding Editor: J. D. Reeve

    work page 2012

  31. [31]

    Cambridge University Press, 2021

    Tom Leinster.Entropy and Diversity: The Axiomatic Approach. Cambridge University Press, 2021

    work page 2021

  32. [32]

    Thedefinitionandmeasurement of heterogeneity.Translational Psychiatry, 10(1):299, 2020

    AbrahamNunes, ThomasTrappenberg, andMartinAlda. Thedefinitionandmeasurement of heterogeneity.Translational Psychiatry, 10(1):299, 2020

    work page 2020

  33. [33]

    Ecological opportunity and the origin of adaptive radiations.Journal of Evolutionary Biology, 23(8):1581–1596, 2010

    JB Yoder, E Clancey, S Des Roches, JM Eastman, L Gentry, W Godsoe, TJ Hagey, D Jochimsen, BP Oswald, JBAJ Robertson, et al. Ecological opportunity and the origin of adaptive radiations.Journal of Evolutionary Biology, 23(8):1581–1596, 2010. 83 The Principle of Maximum Heterogeneity Optimises Productivity in Distributed Production Systems

    work page 2010

  34. [34]

    Biodiversity and ecosystem productivity in a fluctuating environment: The insurance hypothesis.Proceedings of the National Academy of Sciences, 96(4):1463–1468, 1999

    Shigeo Yachi and Michel Loreau. Biodiversity and ecosystem productivity in a fluctuating environment: The insurance hypothesis.Proceedings of the National Academy of Sciences, 96(4):1463–1468, 1999

    work page 1999

  35. [35]

    Biodiversity as spatial insur- ance in heterogeneous landscapes.Proceedings of the National Academy of Sciences, 100(22):12765–12770, 2003

    Michel Loreau, Nicolas Mouquet, and Andrew Gonzalez. Biodiversity as spatial insur- ance in heterogeneous landscapes.Proceedings of the National Academy of Sciences, 100(22):12765–12770, 2003

    work page 2003

  36. [36]

    Population diversity and the portfolio effect in an exploited species.Nature, 465:609–612, 2010

    Daniel E Schindler, Ray Hilborn, Brendan Chasco, et al. Population diversity and the portfolio effect in an exploited species.Nature, 465:609–612, 2010

    work page 2010

  37. [37]

    Shanafelt, Ulf Dieckmann, Matthias Jonas, Oskar Franklin, Michel Loreau, and Charles Perrings

    David W. Shanafelt, Ulf Dieckmann, Matthias Jonas, Oskar Franklin, Michel Loreau, and Charles Perrings. Biodiversity, productivity, and the spatial insurance hypothesis revisited. Journal of Theoretical Biology, 380:426–435, 2015

    work page 2015

  38. [38]

    The human visual cortex.Annual Review of Neuroscience, 27(1):649–677, 2004

    Kalanit Grill-Spector and Rafael Malach. The human visual cortex.Annual Review of Neuroscience, 27(1):649–677, 2004

    work page 2004

  39. [39]

    Receptivefields, binocularinteractionandfunctional architecture in the cat’s visual cortex.The Journal of Physiology, 160(1):106–154, 1962

    DavidHHubelandTorstenNWiesel. Receptivefields, binocularinteractionandfunctional architecture in the cat’s visual cortex.The Journal of Physiology, 160(1):106–154, 1962

    work page 1962

  40. [40]

    A simple coding procedure enhances a neuron’s information capacity

    Simon Laughlin. A simple coding procedure enhances a neuron’s information capacity. Zeitschrift für Naturforschung c, 36(9-10):910–912, 1981

    work page 1981

  41. [41]

    Natural environment statistics in the upper and lower visual field are reflected in mouse retinal specializations.Current Biology, 31(15):3233–3247, 2021

    Yongrong Qiu, Zhijian Zhao, David Klindt, Magdalena Kautzky, Klaudia P Szatko, Frank Schaeffel, Katharina Rifai, Katrin Franke, Laura Busse, and Thomas Euler. Natural environment statistics in the upper and lower visual field are reflected in mouse retinal specializations.Current Biology, 31(15):3233–3247, 2021

    work page 2021

  42. [42]

    Variation in rod and cone density from the fovea to the mid-periphery in healthy human retinas using adaptive optics scanning laser ophthalmoscopy.Eye, 30(8):1135–1143, 2016

    EM Wells-Gray, SS Choi, A Bries, and N Doble. Variation in rod and cone density from the fovea to the mid-periphery in healthy human retinas using adaptive optics scanning laser ophthalmoscopy.Eye, 30(8):1135–1143, 2016

    work page 2016

  43. [43]

    Ashley X Zhou, John Duncan, and Daniel J Mitchell. The default mode subnetworks’ involvement in diverse cognitive transitions suggests a role in external update of internal models.Current Opinion in Behavioral Sciences, 65:101567, 2025

    work page 2025

  44. [44]

    A domain- general cognitive core defined in multimodally parcellated human cortex.Cerebral Cortex, 30(8):4361–4380, 2020

    Moataz Assem, Matthew F Glasser, David C Van Essen, and John Duncan. A domain- general cognitive core defined in multimodally parcellated human cortex.Cerebral Cortex, 30(8):4361–4380, 2020

    work page 2020

  45. [45]

    Construction and use of mental models: Organizing principles for the science of brain and mind.Neuropsychologia, 207:109062, 2025

    John Duncan. Construction and use of mental models: Organizing principles for the science of brain and mind.Neuropsychologia, 207:109062, 2025

    work page 2025

  46. [46]

    Con- scious processing and the global neuronal workspace hypothesis.Neuron, 105(5):776–798, 2020

    George A Mashour, Pieter Roelfsema, Jean-Pierre Changeux, and Stanislas Dehaene. Con- scious processing and the global neuronal workspace hypothesis.Neuron, 105(5):776–798, 2020

    work page 2020

  47. [47]

    The multiple-demand (md) system of the primate brain: mental programs for intelligent behaviour.Trends in cognitive sciences, 14(4):172–179, 2010

    John Duncan. The multiple-demand (md) system of the primate brain: mental programs for intelligent behaviour.Trends in cognitive sciences, 14(4):172–179, 2010

    work page 2010

  48. [48]

    Integrated intelligence from dis- tributed brain activity.Trends in cognitive sciences, 24(10):838–852, 2020

    John Duncan, Moataz Assem, and Sneha Shashidhara. Integrated intelligence from dis- tributed brain activity.Trends in cognitive sciences, 24(10):838–852, 2020

    work page 2020

  49. [49]

    Cambridge Library Collection - British and Irish History, 19th Century

    David Ricardo.On the Principles of Political Economy, and Taxation. Cambridge Library Collection - British and Irish History, 19th Century. Cambridge University Press, 2015. 84 The Principle of Maximum Heterogeneity Optimises Productivity in Distributed Production Systems

    work page 2015

  50. [50]

    Hierarchies and the organization of knowledge in production.Journal of Political Economy, 108(5):874–904, 2000

    Luis Garicano. Hierarchies and the organization of knowledge in production.Journal of Political Economy, 108(5):874–904, 2000

    work page 2000

  51. [51]

    Problemsolvingbyheterogeneousagents.Journal of Economic Theory, 97(1):123–163, 2001

    LuHongandScottE.Page. Problemsolvingbyheterogeneousagents.Journal of Economic Theory, 97(1):123–163, 2001

    work page 2001

  52. [52]

    Lu Hong and Scott E. Page. Groups of diverse problem solvers can outperform groups of high-ability problem solvers.Proceedings of the National Academy of Sciences, 101(46):16385–16389, 2004

    work page 2004

  53. [53]

    Labor specialization and the extent of the market.Journal of Political Economy, 97(3):692–705, 1989

    Sunwoong Kim. Labor specialization and the extent of the market.Journal of Political Economy, 97(3):692–705, 1989

    work page 1989

  54. [54]

    Luis Garicano and Thomas N. Hubbard. Specialization, firms, and markets: The division of labor within and between law firms.Journal of Law, Economics, and Organization, 25(2):339–371, 2009

    work page 2009

  55. [55]

    Portfolio selection.The Journal of Finance, 7(1):77–91, 1952

    Harry Markowitz. Portfolio selection.The Journal of Finance, 7(1):77–91, 1952

    work page 1952

  56. [56]

    Rae, Oriol Vinyals, and Laurent Sifre

    Jordan Hoffmann, Sebastian Borgeaud, Arthur Mensch, Elena Buchatskaya, Trevor Cai, Eliza Rutherford, Diego de Las Casas, Lisa Anne Hendricks, Johannes Welbl, Aidan Clark, Tom Hennigan, Eric Noland, Katie Millican, George van den Driessche, Bogdan Damoc, Aurelia Guy, Simon Osindero, Karen Simonyan, Erich Elsen, Jack W. Rae, Oriol Vinyals, and Laurent Sifre...

    work page 2022

  57. [57]

    Approximation by superpositions of a sigmoidal function.Mathematics of Control, Signals and Systems, 2(4):303–314, 1989

    George Cybenko. Approximation by superpositions of a sigmoidal function.Mathematics of Control, Signals and Systems, 2(4):303–314, 1989

    work page 1989

  58. [58]

    Okoudjou, editor.Finite Frame Theory: A Complete Introduction to Over- completeness, volume 73 ofProceedings of Symposia in Applied Mathematics

    Kasso A. Okoudjou, editor.Finite Frame Theory: A Complete Introduction to Over- completeness, volume 73 ofProceedings of Symposia in Applied Mathematics. American Mathematical Society, Providence, Rhode Island, 2016. AMS Short Course, January 8–9, 2015, San Antonio, Texas

    work page 2016

  59. [59]

    Ronald A. DeVore. Nonlinear approximation.Acta Numerica, 7:51–150, 1998

    work page 1998

  60. [60]

    Nonlinear approximation with dictionaries: Direct estimates.Journal of Approximation Theory, 2018

    Rémi Gribonval and Morten Nielsen. Nonlinear approximation with dictionaries: Direct estimates.Journal of Approximation Theory, 2018

    work page 2018

  61. [61]

    Neural heterogeneity promotes robust learning.Nature communications, 12(1):5791, 2021

    Nicolas Perez-Nieves, Vincent CH Leung, Pier Luigi Dragotti, and Dan FM Goodman. Neural heterogeneity promotes robust learning.Nature communications, 12(1):5791, 2021

    work page 2021

  62. [62]

    Algorithm-hardware co-design of neuromorphic networks with dual memory pathways

    PengfeiSun, ZheSu, JaschaAchterberg, GiacomoIndiveri, DanFMGoodman, andDanyal Akarca. Algorithm-hardware co-design of neuromorphic networks with dual memory path- ways.arXiv preprint arXiv:2512.07602, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  63. [63]

    Scaling Laws for Neural Language Models

    Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling laws for neural language models.arXiv preprint arXiv:2001.08361, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2001

  64. [64]

    Toward maximum diversification.The Journal of Portfolio Management, pages 40–51, 2008

    Yves Choueifaty and Yves Coignard. Toward maximum diversification.The Journal of Portfolio Management, pages 40–51, 2008

    work page 2008

  65. [65]

    Concentration and specialization: Dynamics of niche width in popula- tions of organizations.American Journal of Sociology, 90(6):1262–1283, 1985

    Glenn R Carroll. Concentration and specialization: Dynamics of niche width in popula- tions of organizations.American Journal of Sociology, 90(6):1262–1283, 1985

    work page 1985

  66. [66]

    Why the microbrewery movement? organiza- tional dynamics of resource partitioning in the us brewing industry.American Journal of Sociology, 106(3):715–762, 2000

    Glenn R Carroll and Anand Swaminathan. Why the microbrewery movement? organiza- tional dynamics of resource partitioning in the us brewing industry.American Journal of Sociology, 106(3):715–762, 2000. 85 The Principle of Maximum Heterogeneity Optimises Productivity in Distributed Production Systems

    work page 2000

  67. [67]

    Applications of the principle of maximum entropy: from physics to ecology.Journal of Physics: Condensed Matter, 22(6):063101, 2010

    Jayanth R Banavar, Amos Maritan, and Igor Volkov. Applications of the principle of maximum entropy: from physics to ecology.Journal of Physics: Condensed Matter, 22(6):063101, 2010

    work page 2010

  68. [68]

    NVIDIA vera rubin POD: Seven chips, five rack-scale systems, one AI supercomputer

    Rohil Bhargava, Taylor Allison, and Harry Petty. NVIDIA vera rubin POD: Seven chips, five rack-scale systems, one AI supercomputer. NVIDIA Technical Blog, March 2026. Accessed: 2026-03-26

    work page 2026

  69. [69]

    The hardware lottery.Communications of the ACM, 64(12):58–65, 2021

    Sara Hooker. The hardware lottery.Communications of the ACM, 64(12):58–65, 2021

    work page 2021

  70. [70]

    Asurvey on deep learning hardware accelerators for heterogeneous hpc platforms.ACM Computing Surveys, 57(11):1–39, 2025

    Cristina Silvano, Daniele Ielmini, Fabrizio Ferrandi, Leandro Fiorin, Serena Curzel, Luca Benini, FrancescoConti, AngeloGarofalo, CristianZambelli, EnricoCalore, etal. Asurvey on deep learning hardware accelerators for heterogeneous hpc platforms.ACM Computing Surveys, 57(11):1–39, 2025

    work page 2025

  71. [71]

    Efficient neuromorphic signal processing with loihi 2

    Garrick Orchard, E Paxon Frady, Daniel Ben Dayan Rubin, Sophia Sanborn, Sumit Bam Shrestha, Friedrich T Sommer, and Mike Davies. Efficient neuromorphic signal processing with loihi 2. In2021 IEEE workshop on signal processing systems (SiPS), pages 254–259. IEEE, 2021

    work page 2021

  72. [72]

    TPU v4: An optically reconfigurable supercomputer for machine learning with hardware support for embeddings

    Norm Jouppi, George Kurian, Sheng Li, Peter Ma, Rahul Nagarajan, Lifeng Nai, Nishant Patil, Suvinay Subramanian, Andy Swing, Brian Towles, et al. TPU v4: An optically reconfigurable supercomputer for machine learning with hardware support for embeddings. InProceedings of the 50th annual international symposium on computer architecture, pages 1–14, 2023

    work page 2023

  73. [73]

    Neural inference at the frontier of energy, space, and time.Science, 382(6668):329–335, 2023

    Dharmendra S Modha, Filipp Akopyan, Alexander Andreopoulos, Rathinakumar Ap- puswamy, John V Arthur, Andrew S Cassidy, Pallab Datta, Michael V DeBole, Steven K Esser, Carlos Ortega Otero, et al. Neural inference at the frontier of energy, space, and time.Science, 382(6668):329–335, 2023

    work page 2023

  74. [74]

    The CL1 as a platform technology to leverage biological neural system functions.Nature Reviews Bioengineering, 3(9):724–725, 2025

    Brett J Kagan. The CL1 as a platform technology to leverage biological neural system functions.Nature Reviews Bioengineering, 3(9):724–725, 2025

    work page 2025

  75. [75]

    Richard S. Sutton. The bitter lesson, 2019. Blog post

    work page 2019

  76. [76]

    Athena: Enabling codesign for next- generation ai/ml architectures

    Mark Plagge, Ben Feinberg, John McFarland, Fred Rothganger, Sapan Agarwal, Amro Awad, Clayton Hughes, and Suma G Cardwell. Athena: Enabling codesign for next- generation ai/ml architectures. In2022 IEEE International Conference on Rebooting Computing (ICRC), pages 13–23. IEEE, 2022

    work page 2022

  77. [77]

    Device codesign using re- inforcement learning

    Suma G Cardwell, Karan Patel, Catherine D Schuman, J Darby Smith, Jaesuk Kwon, Andrew Maicke, Jared Arzate, and Jean Anne C Incorvia. Device codesign using re- inforcement learning. In2024 IEEE International Symposium on Circuits and Systems (ISCAS), pages 1–5. IEEE, 2024

    work page 2024

  78. [78]

    A graph placement methodology for fast chip design.Nature, 594(7862):207–212, 2021

    Azalia Mirhoseini, Anna Goldie, Mustafa Yazgan, Joe Wenjie Jiang, Ebrahim Songhori, Shen Wang, Young-Joon Lee, Eric Johnson, Omkar Pathak, Azade Nova, et al. A graph placement methodology for fast chip design.Nature, 594(7862):207–212, 2021

    work page 2021

  79. [79]

    Chipformer: Transferable chip placement via offline decision transformer

    Yao Lai, Jinxin Liu, Zhentao Tang, Bin Wang, Jianye Hao, and Ping Luo. Chipformer: Transferable chip placement via offline decision transformer. InInternational Conference on Machine Learning, pages 18346–18364. PMLR, 2023

    work page 2023

  80. [80]

    Circuitnet 2.0: An advanced dataset for promoting machine learning innovations in realistic chip design environment

    Xun Jiang, Yuxiang Zhao, Yibo Lin, Runsheng Wang, Ru Huang, et al. Circuitnet 2.0: An advanced dataset for promoting machine learning innovations in realistic chip design environment. InThe Twelfth International Conference on Learning Representations, 2024. 86 The Principle of Maximum Heterogeneity Optimises Productivity in Distributed Production Systems

    work page 2024

Showing first 80 references.