Pith. sign in

REVIEW 4 major objections 5 minor 35 references

From Passive Mirrors to Active Agents: Holonic Digital Twins for Physical AI over Networks

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A wireless network of self-reasoning twins can act as one cognitive agent.

desk verdict A clear, well-written vision paper whose central formal claim (Eq. 7) is asserted rather than derived and is undercut by the paper's own Challenge 4; read it as a research agenda. read the letter →

arxiv 2608.06227 v1 pith:R56NNPY7 submitted 2026-08-06 cs.NI cs.AIcs.ITcs.SYeess.SYmath.IT

Christo Kurisummoottil Thomas , Omar Hashash , Walid Saad This is my paper · ORCID
classification cs.NIcs.AIcs.ITcs.SYeess.SYmath.IT
keywords holonicdigitaltwinsactiveinferenceMarkovblanketsexpectedfreeenergyintegratedinformationtheoryphysicalAI6Gnetworkscognitivevalue
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that 6G networks can evolve from passive data pipes into orchestrators of physical AI by pairing every physical agent with a holonic digital twin that actively reasons about the world. The central claim is that a network of such twins, each minimizing its own variational free energy and exchanging beliefs over wireless links, collectively behaves as a single active inference agent that perceives, acts, and learns at a higher abstraction level. If correct, this gives network designers a unified objective for real-time physical AI coordination, replacing throughput-only optimization with decisions driven by the cognitive value of each transmission. The paper also proposes integrated information as a quantitative measure of when coordination produces genuine collective intelligence and how that intelligence grows over a mission.

What carries the argument

The load-bearing object is the blanket of blankets: the outward-facing components $\{s_i^{\mathrm{out}}, a_i^{\mathrm{out}}\}$ of all twins jointly form a collective Markov blanket that screens the aggregate internal beliefs $\mu_{\mathrm{col}}$ from the external environment. Under timescale separation and existence of a collective steady-state density, this blanket structure yields the central gradient-flow equation that turns the composed network into one active-inference agent. Supporting machinery includes expected free energy, which defines the cognitive value of a transmission; categorical pullbacks and natural transformations, which guarantee that heterogeneous world models compose with semantics preserved; and spatiotemporal integrated information, which measures how much of the collective's predictive power is lost when agents are partitioned in space or time.

What would settle it

Simulate or deploy a small network of two or three HDTs with coordination delay comparable to local inference time, then test whether the joint state follows the gradient flow of the paper's equation (7); if the collective trajectory deviates measurably from that flow, or no steady-state density $p(z_{\mathrm{col}})$ can be estimated from repeated runs, the central claim is refuted for that regime.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that collective intelligence is not an added layer over individual agents but a consequence of each twin minimizing its own free energy while coupled through wireless belief exchange. Writing each twin's outward-facing sensory and active states as a collective Markov blanket, the paper shows that when local inference is much faster than coordination and a collective non-equilibrium steady-state density exists, the whole network's autonomous states obey the same gradient flow as a single active inference agent: $\dot{\alpha}_{\mathrm{col}} \propto \nabla_{\alpha_{\mathrm{col}}} \log p(s_{\mathrm{col}}, a_{\mathrm{col}}, \mu_{\mathrm{col}})$. The network is not merely coordinating; it is itself a cognitive agent with its own variational free energy. Beliefs are selected for transmission by their cognitive value, the reduction they induce in the receiver's expected free energy, and the degree of genuine integration is measured by spatiotemporal integrated information $\Phi_{\mathrm{ST}}$.

Load-bearing premise

The central claim rests on two structural conditions: local onboard inference must be much faster than inter-agent wireless coordination, and the collective ensemble must possess a stable non-equilibrium steady-state distribution; if either fails in a real cyber-physical network, the gradient-flow equation that makes the network a single cognitive agent no longer applies.

Editorial extensions

If this is right

  • Coordination topology stops being an engineered optimization: links form, persist, or drop according to whether their mutual information exceeds communication cost, as a side effect of collective free energy minimization.
  • Transmissions are scheduled by cognitive value rather than quality of service: a belief with high Shannon information but no causal relevance to the receiver carries zero value and can be pruned.
  • The network acquires its own free-energy objective, so hierarchical reasoning, stability analysis, and planning can be applied to the collective as though it were a single agent.
  • Resource allocation for sensing, computation, and communication can be aimed at growing spatiotemporal integrated information over mission time, making collective intelligence a managed network resource.
  • Wireless channel quality enters the collective's belief dynamics directly through the coupling $\mu_i \to a_i^{\mathrm{in}} \to \eta_{ij} \to s_j^{\mathrm{in}} \to \mu_j$, so link degradation triggers belief-level recovery actions as part of free energy minimization.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper leaves implicit that spatiotemporal integrated information could serve as a protocol-level feedback signal: a scheduler could periodically estimate $\Phi$ over Markov-blanket neighborhoods and reallocate resources when the measure drops, which is testable in simulation before deployment.
  • If the timescale separation $\tau_{\mathrm{local}} \ll \tau_{\mathrm{coord}}$ fails under dense swarms or congested spectrum, the collective gradient flow becomes an approximation; the paper's framework predicts that coordination overhead should then be treated as a perturbation rather than an exact objective.
  • The cognitive-value definition suggests an information-theoretic generalization of age-of-information: belief staleness should be weighted by the receiver's current epistemic state, so a delayed message during high-uncertainty periods is far costlier than the same delay during confident operation.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes HDT-Net, a framework in which each physical AI agent is paired with a holonic digital twin and wireless networks orchestrate collective physical AI inference. The framework rests on four pillars: causal Markov blankets to partition sensing, communication, and control; category theory to compose heterogeneous world models; active inference to unify perception, action, and learning; and spatiotemporal integrated information to measure collective intelligence. The central formal claim, developed in Section IV-B, is that networked HDTs minimizing their own variational free energy collectively behave as a single active inference agent, expressed by a collective gradient flow (Eq. (7)) and a collective expected free energy decomposition (Eq. (8)). The paper also introduces cognitive value of a transmission (Eq. (4)), a spatiotemporal integration measure (Eq. (9)), and an intelligence growth objective (Eq. (10)). The closing section lists six open challenges, including one that directly limits the central claim.

Significance. If substantiated, the framework would be a valuable conceptual reorientation for 6G research, shifting network design from throughput/latency KPIs toward belief-centric, cognitive objectives. The paper is strong in problem framing: it identifies a genuine gap—wireless networks must provide shared spatiotemporal context for physical AI—and it synthesizes several mathematical tools into a coherent architecture. It is also unusually candid: Challenge 4 explicitly admits that the collective description is coherent only under a synchronization capability that does not yet exist. The strengths are the clarity of the vision, the breadth of the synthesis, and the explicit enumeration of open problems. However, the manuscript is a research manifesto rather than a validated theory: no simulations or case studies are provided, and the central equations are asserted rather than derived. As a result, the paper's intellectual contribution is the architecture and problem framing; the formal results it advertises are not yet established.

major comments (4)
  1. [Section IV-B1, Eq. (7)] The central claim that a network of HDTs collectively behaves as a single active inference agent rests on Eq. (7), a gradient flow stated to hold under conditions (a) and (b). This equation is not derived from the individual variational free energy updates (Eqs. (2)–(3)) and the coupling chain (Eq. (6)); it is imported from an unreviewed preprint [33]. Even granting the timescale separation and the existence of a collective non-equilibrium steady state, Eq. (7) is a gradient flow on a log joint density, whereas individual agents minimize variational free energy F for perception and expected free energy G for action. Bridging those objectives for the collective requires a derivation that is absent. The paper itself concedes in Challenge 4 that the collective description is 'only coherent if the network ensures that agents commit to coordinated actions within a synchronized spatiotemporal context window,' a capability listed as open. Thus the headline claim is presented as a result but is in fact a conjecture conditional on an unresolved problem. I recommend either supplying the derivation or explicitly labeling Eq. (7) as a conjecture or design target.
  2. [Section IV-B2, Eq. (8)] The collective expected free energy decomposition G_col = Σ G_i − Σ I(µ_i; µ_j | s_shared,ij) + C_comm is stated without derivation. It is not shown to follow from a particular joint generative model of the N HDTs, and the sign of the mutual information term is not formally justified; the text suggests that negative mutual information lowers collective free energy, but the relationship between pairwise MI and the local EFE terms needs a derivation. Because the self-organizing topology claim—links are kept when MI exceeds communication cost—depends entirely on this equation, it should either be derived from the joint model or presented as a proposed objective rather than as an established result.
  3. [Section IV-D, Eq. (10)] The intelligence growth objective is internally redundant. The third term, β∫_0^T Φ̇(t)dt, equals β(Φ(T)−Φ(0)). Since Φ(0) is a constant for a given initial condition, this term is linearly dependent on the terminal intelligence term αΦ(T); the two weights α and β are not independently identifiable, and the objective cannot actually balance 'terminal intelligence' against 'intelligence growth.' A genuinely distinct growth term—for example, an integral with a time-dependent weighting that does not telescope—is needed to realize the stated design goal.
  4. [Section III-B2 and III-C4] The paper claims that category theory provides 'formal guarantees' that semantic structure is preserved across heterogeneous agents, but the existence of the functors Φ_DT, Φ_abs, Φ_ToM and of the natural transformation α is assumed rather than established. For arbitrary pairs of world-model implementations, a structure-preserving translation may not exist; the claimed guarantee is therefore conditional on an existence assumption that is never stated or checked. The pullback construction in Fig. 2 is well defined as a composition operation, but the semantic-consistency guarantee goes beyond the pullback and needs a separate argument or an explicit existence condition.
minor comments (5)
  1. [Section IV-A, near Eq. (2)] The sentence explaining the complexity term writes E_Q[ln Q(µ_t) / ln P(µ_t|...)]; this should be E_Q[ln Q(µ_t) − ln P(µ_t|...)], since the KL divergence is the expectation of a log-ratio, not a ratio of logarithms.
  2. [Section II-B4] There are spacing typos such as '90percent' and 'agent3trajectory' that should be corrected to '90 percent' and 'agent 3 trajectory'.
  3. [Section IV-C, Eq. (9)] The admissible set of partitions P_ST(R,Δt) is not defined rigorously; in particular, it is unclear whether partitions must respect spatial contiguity or temporal-window constraints, and how the minimization over partitions operationalizes the intuitive statement about losing temporal synchronization should be clarified.
  4. [Section IV-B1] The definition of the collective autonomous states α_col = (µ_col, a_col) is ambiguous: a_col is not explicitly defined in terms of the inward/outward decompositions introduced just above, whereas µ_col is explicitly given as {µ_i, s_in_i, a_in_i}.
  5. [References [32], [33]] Key mathematical support (especially Eq. (7)) is drawn from an unreviewed preprint [33] and an in-preparation manuscript [32]; these sources should be either replaced by peer-reviewed derivations or clearly marked as auxiliary, since the paper's central claim depends on them.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular reduction found: the collective-agent claim rests on an imported gradient-flow assumption (Eq. 7) under explicit conditions, not on a fitted input or self-citation chain; self-citations are minor and non-load-bearing.

full rationale

HDT-Net is a synthesis/vision paper. Its main quantitative objects are definitions or standard expressions: variational free energy (Eq. 2), expected free energy (Eq. 3), cognitive value (Eqs. 4-5), spatiotemporal integrated information (Eq. 9), and the growth objective (Eq. 10). No parameter is fitted to a subset of data and then renamed as a prediction, and no equation is shown to equal another by construction. The central claim that the network is itself a cognitive agent is formally supported only by Eq. (7), which is introduced by citation to an external preprint [33] under conditions (a)-(b); it is not derived from the individual VFE updates in Eqs. (2)-(3). That is an underived, load-bearing assumption, but it is not circular by construction and it is not forced by the authors' own prior results. The self-citations [32] and [35] appear as motivation or feasibility evidence and are not the mechanism that establishes Eq. (7). Challenge 4 concedes that the synchronized temporal context needed for Eq. (7) is an open problem, which weakens the support for the headline claim but does not make the derivation circular. Overall, no specific reduction from output to input, no fitted-parameter prediction, and no load-bearing self-citation chain was found; a score of 2 reflects only the presence of minor self-citations and unpublished references, not circularity.

Assumptions & free parameters 2 free parameters · 6 assumptions · 4 invented entities

The central claims rest on several unverified modeling assumptions, most importantly the existence of a collective steady-state density and the validity of the gradient-flow description. The framework introduces new constructs such as HDT, cognitive value, and Phi_ST, none of which are supported by independent evidence in this paper. The two objective weights alpha and beta are free parameters chosen by the system designer.

free parameters (2)
  • alpha (intelligence level weight)
    Introduced in objective J in Eq. (10) to balance immediate task reward against terminal integrated information; no fitting procedure is given, it is a hand-chosen mission-specific weight.
  • beta (intelligence growth weight)
    Introduced in Eq. (10) to weight the integral of Phi-dot over the mission horizon; chosen by the designer, not fitted to data.
assumptions (6)
  • domain assumption Existence of a collective non-equilibrium steady-state density p(z_col) for the ensemble of HDTs.
    Stated as condition (b) in Section IV-B1; the collective gradient flow in Eq. (7) requires this density to exist and be differentiable.
  • domain assumption Timescale separation tau_local << tau_coord between onboard inference and wireless coordination.
    Condition (a) in Section IV-B1; needed for the blanket-of-blankets to screen internal from external states and for the gradient-flow description to be valid.
  • domain assumption Gradient-flow characterization of active inference agents as in [33].
    Eq. (7) is lifted from an unreviewed engineering preprint and assumed to hold for each HDT and for the collective.
  • domain assumption Causal Markov blankets can be dynamically detected from conditional mutual information in cyber-physical networks.
    Section III-A relies on this to justify adaptive blanket boundaries; the paper cites [31] without demonstrating convergence or practicality.
  • standard math Natural transformations between time-indexed functors preserve semantic meaning across heterogeneous representations.
    The categorical machinery is standard, but its application to semantic interoperability of world models is an assumption, not a theorem in this paper.
  • domain assumption Integrated information theory provides a valid measure of collective intelligence.
    The paper adopts IIT's Phi as the network-level coordination metric, including the novel spatiotemporal decomposition, without empirical validation.
invented entities (4)
  • Holonic digital twin (HDT)
    purpose: An active cognitive twin paired with each physical AI agent, reasoning autonomously and composing into larger cognitive units.
    No experimental or simulation evidence is provided; it is the central proposed abstraction.
  • Cognitive value of a transmission V(m_i->j)
    purpose: Quantifies the reduction in the receiver's expected free energy to prioritize belief exchanges.
    Defined via Eq. (4) but never measured or computed in any scenario.
  • Spatiotemporal integrated information Phi_ST
    purpose: Measures when a network of HDTs operates as a genuinely integrated collective rather than independent agents.
    Defined via Eq. (9) with a partition set, but no algorithm for its computation at network scale is given.
  • Theory-of-mind functor Phi_ToM
    purpose: Maps one agent's beliefs to a prediction of another agent's beliefs, enabling cognitive value computation.
    Introduced in Section III-C2; no constructive learning rule is provided, only reference to prior work [5].

how reviews work

0 comments
Cite this review

Pith. "Pith review of From Passive Mirrors to Active Agents: Holonic Digital Twins for Physical AI over Networks." pith.science (2026). https://pith.science/paper/R56NNPY7

@misc{pith2026260806227,
  author       = {Pith},
  title        = {Pith review of: From Passive Mirrors to Active Agents: Holonic Digital Twins for Physical AI over Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/R56NNPY7}},
  note         = {Machine review of arXiv:2608.06227}
}
read the original abstract

Despite advances in artificial intelligence (AI) across multiple sectors, today's AI tools, including deep learning and generative AI, still fail when embedded into physical systems, such as robots and vehicles operating under real-world physical laws. This stems from their inability to maintain reliable world models for long-horizon planning under uncertainty and generalize to unseen scenarios. In this context, wireless networks, through pervasive sensing and communication, can orchestrate physical intelligence. However, current architectures optimize throughput, latency, and reliability and cannot support real-time physical AI coordination, requiring agents to maintain shared spatiotemporal context. To address these challenges, a network of holonic digital twins (HDT-Nets) framework is proposed to deliver real-time physical AI inference through holonic agents that actively reason about their environment rather than passively mirror physical assets. Each HDT is realized as a hierarchical structure spanning the physical agent and network edge, reasoning autonomously at the local level while cooperating with neighboring HDTs to form collectively intelligent units. In HDT-Net, causal Markov blankets spanning sensing, communication, and control determine which agents must coordinate and enable counterfactual reasoning over multi-domain interventions. Active inference within these boundaries unifies perception, action, and learning by minimizing expected free energy while deciding which beliefs to transmit based on their cognitive value to the receiver. Category theory ensures that transmitted beliefs preserve semantic structure across heterogeneous agents with incompatible representations. Finally, integrated information theory quantifies when collective intelligence exceeds independent operation and how network intelligence evolves through coordinated learning and information exchange.

Figures

Figures reproduced from arXiv: 2608.06227 by the authors.

Figure 1
Figure 1. Overview of Proposed HDT-Net. all scenarios the physical AI network might encounter. But physical AI environments are inherently non-stationary and may encounter unexpected scenarios that may occur due to emergent interactions among the agents. No historical dataset can anticipate all futures. Moreover, neural models extrapolate poorly beyond their training distribution since they learn purely statistical associatio… view at source ↗
Figure 2
Figure 2. Pullback diagram for composing digital twins. The composite DT1,2 projects to individual DTs via π1 and π2, which then map to shared state via f1 and f2. The pullback property ensures f1 ◦ π1 = f2 ◦ π2, maintaining consistency on shared components. tees that independently developed HDTs compose through standardized wireless interfaces with well-defined behavior rather than undefined interactions. Functorial abstract… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 24 canonical work pages

  1. [33]

    Active inference for physical ai agents: An engineering perspective,

    B. de Vries, “Active inference for physical ai agents: An engineering perspective,”arXiv preprint arXiv:2603.20927, 2026

    arXiv 2026

  2. [1]

    Artificial general intelligence (AGI)-native wireless systems: A journey beyond 6G,

    W. Saad, O. Hashash, C. K. Thomas, C. Chaccour, M. Debbah, N. Man- dayam, and Z. Han, “Artificial general intelligence (AGI)-native wireless systems: A journey beyond 6G,”Proceedings of the IEEE, 2025

    work page 2025

  3. [2]

    Causal reasoning: Charting a revolutionary course for next-generation AI-native wireless networks,

    C. K. Thomas, C. Chaccour, W. Saad, M. Debbah, and C. S. Hong, “Causal reasoning: Charting a revolutionary course for next-generation AI-native wireless networks,”IEEE Vehicular Technology Magazine, vol. 19, no. 1, Mar. 2024

    work page 2024

  4. [3]

    Time, simultaneity, and causality in wireless networks with sensing and communications,

    P. Popovski, “Time, simultaneity, and causality in wireless networks with sensing and communications,”IEEE Open Journal of the Communica- tions Society, vol. 5, pp. 1693–1709, 2024

    work page 2024

  5. [4]

    Information value theory,

    R. A. Howard, “Information value theory,”IEEE Transactions on Systems Science and Cybernetics, vol. 2, no. 1, pp. 22–26, 1966

    1966

  6. [5]

    Metamind: General and cognitive world models in multi-agent systems by meta-theory of mind,

    L. Wang, R. Shelim, W. Saad, and N. Ramakrishna, “Metamind: General and cognitive world models in multi-agent systems by meta-theory of mind,”arXiv preprint arXiv:2603.00808, 2026

    arXiv 2026

  7. [6]

    Dmwm: Dual- mind world model with long-term imagination,

    L. Wang, R. Shelim, W. Saad, and N. Ramakrishnan, “Dmwm: Dual- mind world model with long-term imagination,”Advances in Neural Information Processing Systems, vol. 38, pp. 4911–4945, 2026

    work page 2026

  8. [7]

    Digital-twin- enabled 6G: Vision, architectural trends, and future directions,

    L. U. Khan, W. Saad, D. Niyato, Z. Han, and C. S. Hong, “Digital-twin- enabled 6G: Vision, architectural trends, and future directions,”IEEE Communications Magazine, vol. 60, no. 1, pp. 74–80, Jan. 2022

    2022

Show all 35 references
  1. [8]

    A path towards autonomous machine intelligence,

    Y . LeCun, “A path towards autonomous machine intelligence,”OpenRe- view, vol. 62, no. 1, 2022, version 0.9.2, 2022-06-27

    2022

  2. [9]

    From simulators to digital twins for enabling emerging cellular networks: A tutorial and survey,

    M. Manalastas, M. U. B. Farooq, S. M. A. Zaidi, H. N. Qureshi, Y . Sambo, and A. Imran, “From simulators to digital twins for enabling emerging cellular networks: A tutorial and survey,”IEEE Communica- tions Surveys & Tutorials, vol. 27, no. 4, pp. 2693–2732, 2024

    2024

  3. [10]

    Digital twin for open RAN: Toward intelligent and resilient 6G radio access networks,

    A. Masaracchia, V . Sharma, M. Fahim, O. A. Dobre, and T. Q. Duong, “Digital twin for open RAN: Toward intelligent and resilient 6G radio access networks,”IEEE Communications Magazine, vol. 61, no. 11, pp. 112–118, 2023

    2023

  4. [11]

    Digital twins and multi-access edge computing for IIoT,

    A. P. Plageras and K. E. Psannis, “Digital twins and multi-access edge computing for IIoT,”Virtual Reality & Intelligent Hardware, vol. 4, no. 6, pp. 521–534, 2022

    2022

  5. [12]

    Digital-twin-aided dynamic spectrum sharing and resource management in integrated satellite-terrestrial networks,

    H. Nguyen-Kha, V . N. Ha, T. Nguyen, E. Lagunas, J. Grotz, S. Chatzino- tas, and B. Ottersten, “Digital-twin-aided dynamic spectrum sharing and resource management in integrated satellite-terrestrial networks,”arXiv preprint arXiv:2602.09191, 2026

    2026 arXiv

  6. [13]

    Generative ai empowered network digital twins: Architecture, technologies, and applications,

    T. Li, Q. Long, H. Chai, S. Zhang, F. Jiang, H. Liu, W. Huang, D. Jin, and Y . Li, “Generative ai empowered network digital twins: Architecture, technologies, and applications,”ACM Computing Surveys, vol. 57, no. 6, pp. 1–43, 2025

    2025

  7. [14]

    Llm-integrated digital twins for hierarchical resource allocation in 6g networks,

    M. Haider, I. Ahmed, Z. Hassan, K. Hasan, and H. Poor, “Llm-integrated digital twins for hierarchical resource allocation in 6g networks,”arXiv preprint arXiv:2506.18293, 2025

    2025 arXiv

  8. [15]

    Learning radio environments by dif- ferentiable ray tracing,

    J. Hoydis, F. A. Aoudia, S. Cammerer, F. Euchner, M. Nimier-David, S. T. Brink, and A. Keller, “Learning radio environments by dif- ferentiable ray tracing,”IEEE Transactions on Machine Learning in Communications and Networking, vol. 2, pp. 1527–1539, 2024

    2024

  9. [16]

    Large wireless model (lwm): A foundation model for wireless channels,

    S. Alikhani, G. Charan, and A. Alkhateeb, “Large wireless model (lwm): A foundation model for wireless channels,”arXiv preprint arXiv:2411.08872, 2024

    2024 arXiv

  10. [17]

    Koestler,The Ghost in the Machine, 1968

    A. Koestler,The Ghost in the Machine, 1968

    1968

  11. [18]

    H. A. Simon,The Sciences of the Artificial, 3rd ed. Cambridge, MA: MIT Press, 1996

    1996

  12. [19]

    Fundamentals of holonic systems and their implications for self-adaptive and self-organizing systems,

    P. Valckenaers, H. V . Brussel, and T. Holvoet, “Fundamentals of holonic systems and their implications for self-adaptive and self-organizing systems,” inProceedings of the Second IEEE International Conference on Self-Adaptive and Self-Organizing Systems Workshops, 2008, pp. 168–173

    2008

  13. [20]

    Using markov blankets for causal structure learning,

    J.-P. Pellet and A. Elisseeff, “Using markov blankets for causal structure learning,”Journal of Machine Learning Research, vol. 9, no. 7, 2008

    2008

  14. [21]

    Barr and C

    M. Barr and C. Wells,Category theory for computing science. Prentice Hall New York, 1990, vol. 1

    1990

  15. [22]

    Active inference: a process theory,

    K. Friston, T. FitzGerald, F. Rigoli, P. Schwartenbeck, and G. Pezzulo, “Active inference: a process theory,”Neural computation, vol. 29, no. 1, pp. 1–49, 2017

    2017

  16. [23]

    Integrated information theory: from consciousness to its physical substrate,

    G. Tononi, M. Boly, M. Massimini, and C. Koch, “Integrated information theory: from consciousness to its physical substrate,”Nature Reviews Neuroscience, vol. 17, no. 7, pp. 450–461, 2016

    2016

  17. [24]

    Kahneman,Thinking, fast and slow

    D. Kahneman,Thinking, fast and slow. macmillan, 2011

    2011

  18. [25]

    Deep learning on multimodal sensor data at the wireless edge for vehicular network,

    B. Salehi, G. Reus-Muns, D. Roy, Z. Wang, T. Jian, J. Dy, S. Ioannidis, and K. Chowdhury, “Deep learning on multimodal sensor data at the wireless edge for vehicular network,”IEEE Transactions on Vehicular Technology, vol. 71, no. 7, pp. 7639–7655, 2022

    2022

  19. [26]

    Deep learning for hybrid 5G services in mobile edge computing systems: Learn from a digital twin,

    R. Dong, C. She, W. Hardjawana, Y . Li, and B. Vucetic, “Deep learning for hybrid 5G services in mobile edge computing systems: Learn from a digital twin,”IEEE Transactions on Wireless Communications, vol. 18, no. 10, pp. 4692–4707, 2019

    2019

  20. [27]

    Toward intelligent cyber-physical systems: Digital twin meets artificial intelligence,

    M. Groshev, C. Guimar ˜aes, J. Mart ´ın-P´e´erez, and A. de la Oliva, “Toward intelligent cyber-physical systems: Digital twin meets artificial intelligence,”IEEE Communications Magazine, vol. 59, no. 8, pp. 14– 20, 2021

    2021

  21. [28]

    Multi-agent collaboration mechanisms: A survey of llms,

    K. T. Tran, D. Dao, M. D. Nguyen, Q. V . Pham, B. O’Sullivan, and H. D. Nguyen, “Multi-agent collaboration mechanisms: A survey of llms,”arXiv preprint arXiv:2501.06322, 2025

    2025 arXiv

  22. [29]

    Aleatoric and epistemic uncertainty in machine learning: An introduction to concepts and methods,

    E. H ¨ullermeier and W. Waegeman, “Aleatoric and epistemic uncertainty in machine learning: An introduction to concepts and methods,”Machine Learning, vol. 110, no. 3, pp. 457–506, 2021

    2021

  23. [30]

    Cooperative heterogeneous multi- robot systems: A survey,

    Y . Rizk, M. Awad, and E. W. Tunstel, “Cooperative heterogeneous multi- robot systems: A survey,”ACM Computing Surveys (CSUR), vol. 52, no. 2, pp. 1–31, 2019

    2019

  24. [31]

    Dynamic markov blanket detection for macroscopic physics discovery,

    J. Beck and M. J. D. Ramstead, “Dynamic markov blanket detection for macroscopic physics discovery,”arXiv preprint arXiv:2502.21217, Feb 2025

    2025 arXiv

  25. [32]

    Active inference as the test-time scaling law for physical AI agents,

    O. Hashash, C. K. Thomas, W. Saad, M. Debbah, K. Friston, and A. Razi, “Active inference as the test-time scaling law for physical AI agents,” 2026, in preparation. 16

    2026

  26. [34]

    Age of information: A new concept, metric, and tool,

    A. Kosta, N. Pappas, and V . Angelakis, “Age of information: A new concept, metric, and tool,”Foundations and Trends in Networking, vol. 12, no. 3, pp. 162–259, 2017

    2017

  27. [35]

    Causal Semantic Communication for Digital Twins: A Generalizable Imitation Learning Approach,

    C. K. Thomas, W. Saad, and Y . Xiao, “Causal Semantic Communication for Digital Twins: A Generalizable Imitation Learning Approach,”IEEE Journals on Selected Areas in Information Theory, vol. 4, pp. 698–717, Nov. 2023

    2023

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.