arXiv subjects

Fernando E. Rosas

Publications and source records attributed to Fernando E. Rosas.

At least 19 recordsLinked to original sources

arXiv2026

World models of environment, agent and joint agent-environment systems

World models are a central component of model-based reinforcement learning. They are usually discussed in terms of what variables they predict, such as observations, rewards, states, latent or information states. We argue that there is a prior distinction: which channel they model. We consider three cases: the environment channel $O_{:} \mid A_{:}$, the agent channel $A_{:} \mid O_{:}$, and the realised joint process $(A, O)_{:}$, equivalently viewed as a channel with no inputs. Using computational mechanics, we define canonical predictive models for these three cases as $ε$-transducers or $ε$-machines. Canonical environment models recover standard predictive state representations, while the other two give analogous notions of canonical models for the agent and the joint system. We then build canonical support-restricted environment and agent models induced by closed-loop coupling, whose predictive equivalences range over continuations supported by the realised interaction. The key structural result is that canonical support-restricted environment states factor through the canonical joint causal states, and their transition structure is induced directly from the joint model; the agent-side construction is dual. Finally, we give a POMDP/controller example in which the unrestricted environment model has infinitely many states while the canonical support-restricted model induced by the coupling is finite. The framework clarifies what different world models are models of, and how coupling and support restriction can change their canonical predictive structure and complexity.

cs.AI↗

arXiv2026

Generalised Bellman recurrence and three dualities in sequential decision-making

What gives the Bellman equation its form? We show that the recursive properties of optimal value functions follow from three conditions: that the dynamics decomposes through sufficient statistics, that the return decomposes recursively, and that the aggregation of uncertainty is compatible with both. When all three conditions hold on a common state, the Bellman equation arises from their mutual consistency; when one fails, tractability can often be recovered by augmenting the state or by deforming return or dynamics. The same conditions are shown to give rise to three dualities: one between probability and return, one between return and aggregation, and one between aggregation and probability. Our framework reveals these dualities as arising from a single construction, unifying methods developed separately across reinforcement learning, control, and decision theory.

cs.LG↗

arXiv2026

Hide and Seek in Embedding Space: Geometry-based Steganography and Detection in Large Language Models

Fine-tuned LLMs can covertly encode prompt secrets into outputs via steganographic channels. Prior work demonstrated this threat but relied on trivially recoverable encodings. We formalize payload recoverability via classifier accuracy and show previous schemes achieve 100\% recoverability. In response, we introduce low-recoverability steganography, replacing arbitrary mappings with embedding-space-derived ones. For Llama-8B (LoRA) and Ministral-8B (LoRA) trained on TrojanStego prompts, exact secret recovery rises from 17$\rightarrow$30\% (+78\%) and 24$\rightarrow$43\% (+80\%) respectively, while on Llama-70B (LoRA) trained on Wiki prompts, it climbs from 9$\rightarrow$19\% (+123\%), all while reducing payload recoverability. We then discuss detection. We argue that detecting fine-tuning-based steganographic attacks requires approaches beyond traditional steganalysis. Standard approaches measure distributional shift, which is an expected side-effect of fine-tuning. Instead, we propose a mechanistic interpretability approach: linear probes trained on later-layer activations detect the secret with up to 33\% higher accuracy in fine-tuned models compared to base models, even for low-recoverability schemes. This suggests that malicious fine-tuning leaves actionable internal signatures amenable to interpretability-based defenses.

cs.CR↗

arXiv2026

Now You (Still) See Me: Detecting Evasive Steganographic Payloads in LLMs

Large language models can be fine-tuned to encode prompt-borne secrets into fluent, seemingly benign outputs. This creates a steganographic exfiltration risk that is difficult to detect with output-level steganalysis. Recent work proposes mechanistic detection using linear probes that recover the secret from internal activations. We show that this defense can be systematically evaded, but that detectability can be recovered through a targeted data-level intervention. First, we extend the detection setup to include a non-linear MLP probe. We then adversarially fine-tune steganographic trojans across five base models: Qwen3-8B, Llama-3.1-8B, Ministral-8B, Qwen3-14B, and Phi-4-14B. The resulting models retain $58$--$79\%$ exact-match secret recovery while evading both ridge and held-out MLP probes, with $1$--$8\%$ average capability degradation across six benchmarks. We then give an information-theoretic characterization of this evasion. Successful evasion preserves recoverability while reducing low-order extractability of the secret from the content-aligned representation, forcing the payload into synergistic interaction with residual degrees of freedom. This motivates a recontextualization dataset that restricts these residual degrees of freedom. On this distribution, both ridge and MLP detectability are restored across all five evasive trojans. Overall, our findings show that activation-based steganography detection is vulnerable to adaptive evasion, but also that theory-guided evaluation distributions can expose otherwise hidden payloads.

cs.CR↗

arXiv2026

Adaptive state-action abstractions via rate-distortion

When learning to walk, infants seem to address a coarse version of the problem first - stay upright, reach the caregiver - and refine it only when further practice at that resolution stops paying off. Reinforcement learning offers multiple techniques for building simple versions of complex tasks, but lacks general principles for how to dynamically adjust the granularity of these abstractions during learning. This paper proposes one such principle: refine the abstraction as soon as the learning error within it becomes comparable to the error induced by the abstraction itself. Here, we investigate one way of formalising this principle via a performance certificate that decomposes value error into two terms: a learning error bound captured by a Bellman residual, and an abstraction error bound given by a bisimulation metric. The resulting switching strategy is implemented by soft state-action abstractions built from rate-distortion principles, whose resolution along state and action axes can be continuously adjusted. We validate this construction in a range of tabular settings, showing that near-optimal performance can be achieved under substantial lossy compression of state and action information.

cs.LG↗

arXiv2026

From monoliths to modules: Decomposing transducers for efficient world modelling

World models have been recently proposed as sandbox environments in which AI agents can be trained and evaluated before deployment. While realistic world models often have high computational demands, this can often be alleviated by exploiting the fact that real-world scenarios tend to involve subcomponents that interact in a modular manner. In this paper, we explore this idea by developing a framework for decomposing complex world models represented by transducers, a class of models generalising POMDPs. Whereas the composition of transducers is well understood, our results clarify how to invert this process by deriving sub-transducers operating on distinct input-output subspaces, enabling parallelizable and interpretable alternatives to monolithic world modelling that can support distributed inference. Overall, these results lay groundwork for bridging the computational efficiency required for real-world inference and the structural transparency demanded by AI safety.

cs.AI↗

arXiv2026

Integrated information theory: the good, the bad and the misunderstood

The integrated information theory of consciousness (IIT) is uniquely ambitious in proposing a mathematical formula, derived from apparently fundamental properties of conscious experience, to describe the quantity and quality of consciousness for any physical system that possesses it. IIT has generated considerable debate, which has engendered some misunderstandings and misrepresentations. Here we address and hope to remedy this. We begin by concisely summarising the essentials of IIT. Given IIT is supposed to apply universally, we do this with reference to an arbitrary patch of matter, as opposed to the usual system of discrete computational units. Then, after briefly summarising IIT's theoretical and empirical achievements, we focus on five points which we consider especially important for driving forward new theory and increasing understanding. First, a high value of the measure $Φ$ is not synonymous with `more consciousness'. We describe how $Φ$ might be replaced with a suite of quantities to obtain a multi-dimensional characterisation of states of consciousness. Second, we describe with nuance the distinct flavour of panpsychism implied by IIT -- whereby space (and time) are tiled with substrates of (proto-) consciousness -- and find this is not problematic for the theory. Third, $Φ$ is not well-defined for real physical systems, and has not been computed on any real physical system. Fourth, so far only proxies for IIT measures have been computed, and not approximations. Fifth, for IIT to fit with current successful theories in fundamental physics, a reformulation in terms of continuous fields would be needed.

q-bio.NC↗

arXiv2026

A scalable estimator of higher-order information in complex dynamical systems

Our understanding of complex systems rests on our ability to characterise how they perform distributed computation and integrate information. Advances in information theory have introduced several quantities to describe complex information structures, where collective patterns of coordination emerge from higher-order (i.e. beyond-pairwise) interdependencies. Unfortunately, the use of these approaches to study large complex systems is severely hindered by the poor scalability of existing techniques. Moreover, there are relatively few measures specifically designed for multivariate time series data. Here we introduce a novel measure of information about macroscopic structures, termed M-information, which quantifies the higher-order integration of information in complex dynamical systems. We show that M-information can be calculated via a convex optimisation problem, and we derive a robust and efficient algorithm that scales gracefully with system size. Our analyses show that M-information is resilient to noise, indexes critical behaviour in artificial neuronal populations, and reflects states of consciousness and task performance in real-world macaque and mouse neuroimaging data. Furthermore, M-information can be incorporated into existing information decomposition frameworks to reveal a comprehensive taxonomy of information dynamics. Taken together, these results help us unravel collective computation in large complex systems.

cs.IT↗

arXiv2026

Towards Worst-Case Guarantees with Scale-Aware Interpretability

Neural networks organize information according to the hierarchical, multi-scale structure of natural data. Methods to interpret model internals should be similarly scale-aware, explicitly tracking how features compose across resolutions and guaranteeing bounds on the influence of fine-grained structure that is discarded as irrelevant noise. We posit that the renormalisation framework from physics can meet this need by offering technical tools that can overcome limitations of current methods. Moreover, relevant work from adjacent fields has now matured to a point where scattered research threads can be synthesized into practical, theory-informed tools. To combine these threads in an AI safety context, we propose a unifying research agenda -- \emph{scale-aware interpretability} -- to develop formal machinery and interpretability tools that have robustness and faithfulness properties supported by statistical physics.

hep-th↗

arXiv2026

Transformers learn factored representations

Transformers pretrained via next token prediction learn to factor their world into parts, representing these factors in orthogonal subspaces of the residual stream. We formalize two representational hypotheses: (1) a representation in the product space of all factors, whose dimension grows exponentially with the number of parts, or (2) a factored representation in orthogonal subspaces, whose dimension grows linearly. The factored representation is lossless when factors are conditionally independent, but sacrifices predictive fidelity otherwise, creating a tradeoff between dimensional efficiency and accuracy. We derive precise predictions about the geometric structure of activations for each, including the number of subspaces, their dimensionality, and the arrangement of context embeddings within them. We test between these hypotheses on transformers trained on synthetic processes with known latent structure. Models learn factored representations when factors are conditionally independent, and continue to favor them early in training even when noise or hidden dependencies undermine conditional independence, reflecting an inductive bias toward factoring at the cost of fidelity. This provides a principled explanation for why transformers decompose the world into parts, and suggests that interpretable low dimensional structure may persist even in models trained on complex data.

cs.LG↗

arXiv2026

A Brain-like Synergistic Core in LLMs Drives Behaviour and Learning

The independent evolution of intelligence in biological and artificial systems offers a unique opportunity to identify its fundamental computational principles. Here we show that large language models spontaneously develop synergistic cores -- components where information integration exceeds individual parts -- remarkably similar to those in the human brain. Using principles of information decomposition across multiple LLM model families and architectures, we find that areas in middle layers exhibit synergistic processing while early and late layers rely on redundancy, mirroring the informational organisation in biological brains. This organisation emerges through learning and is absent in randomly initialised networks. Crucially, ablating synergistic components causes disproportionate behavioural changes and performance loss, aligning with theoretical predictions about the fragility of synergy. Moreover, fine-tuning synergistic regions through reinforcement learning yields significantly greater performance gains than training redundant components, yet supervised fine-tuning shows no such advantage. This convergence suggests that synergistic information processing is a fundamental property of intelligence, providing targets for principled model design and testable predictions for biological intelligence.

cs.AI↗

arXiv2025

Improved estimators of causal emergence for large systems

A central challenge in the study of complex systems is the quantification of emergence -- understood as the ability of the system to exhibit collective behaviours that cannot be traced down to the individual components. While recent work has proposed practical measures to detect emergence, these approaches tend to double-count the contribution of shared components, which substantially hinders their capability to effectively study large systems. In this work, we introduce a family of improved information-theoretic measures of emergence that iteratively correct for double-counted terms. Our approach is computationally efficient and provides a controllable trade-off between computational load and sensitivity, leading to more accurate and versatile estimates of emergence. The benefits of the proposed approach are demonstrated by successfully detecting emergence in both simulated and real-world data related to flocking behaviour.

physics.soc-ph↗

arXiv2025

Explosive neural networks via higher-order interactions in curved statistical manifolds

Higher-order interactions underlie complex phenomena in systems such as biological and artificial neural networks, but their study is challenging due to the scarcity of tractable models. By leveraging a generalisation of the maximum entropy principle, we introduce curved neural networks as a class of models with a limited number of parameters that are particularly well-suited for studying higher-order phenomena. Through exact mean-field descriptions, we show that these curved neural networks implement a self-regulating annealing process that can accelerate memory retrieval, leading to explosive order-disorder phase transitions with multi-stability and hysteresis effects. Moreover, by analytically exploring their memory-retrieval capacity using the replica trick, we demonstrate that these networks can enhance memory capacity and robustness of retrieval over classical associative-memory networks. Overall, the proposed framework provides parsimonious models amenable to analytical study, revealing higher-order phenomena in complex networks.

cond-mat.dis-nn↗

arXiv2025

Symmetries at the origin of hierarchical emergence

Many systems of interest exhibit nested emergent layers with their own rules and regularities, and our knowledge about them seems naturally organised around these levels. This paper proposes that this type of hierarchical emergence arises as a result of underlying symmetries. By combining principles from information theory, group theory, and statistical mechanics, one finds that dynamical processes that are equivariant with respect to a symmetry group give rise to emergent macroscopic levels organised into a hierarchy determined by the subgroups of the symmetry. The same symmetries happen to also shape Bayesian beliefs, yielding hierarchies of abstract belief states that can be updated autonomously at different levels of resolution. These results are illustrated in Hopfield networks and Ehrenfest diffusion, showing that familiar macroscopic quantities emerge naturally from their symmetries. Together, these results suggest that symmetries provide a fundamental mechanism for emergence and support a structural correspondence between objective and epistemic processes, making feasible inferential problems that would otherwise be computationally intractable.

q-bio.NC↗

arXiv2025

Shannon invariants: A scalable approach to information decomposition

Distributed systems, such as biological and artificial neural networks, process information via complex interactions engaging multiple subsystems, resulting in high-order patterns with distinct properties across scales. Investigating how these systems process information remains challenging due to difficulties in defining appropriate multivariate metrics and ensuring their scalability to large systems. To address these challenges, we introduce a novel framework based on what we call "Shannon invariants" -- quantities that capture essential properties of high-order information processing in a way that depends only on the definition of entropy and can be efficiently calculated for large systems. Our theoretical results demonstrate how Shannon invariants can be used to resolve long-standing ambiguities regarding the interpretation of widely used multivariate information-theoretic measures. Moreover, our practical results reveal distinctive information-processing signatures of various deep learning architectures across layers, which lead to new insights into how these systems process information and how this evolves during training. Overall, our framework resolves fundamental limitations in analyzing high-order phenomena and offers broad opportunities for theoretical developments and empirical analyses.

cs.IT↗

arXiv2025

Thermodynamics of exponential Kolmogorov-Nagumo averages

This paper investigates generalized thermodynamic relationships in physical systems where relevant macroscopic variables are determined by the exponential Kolmogorov-Nagumo average. We show that while the thermodynamic entropy of such systems is naturally described by Rényi's entropy with parameter $γ$, an ordinary Boltzmann distribution still describes their statistics under equilibrium thermodynamics. Our results show that systems described by exponential Kolmogorov-Nagumo averages can be interpreted as systems originally in thermal equilibrium with a heat reservoir with inverse temperature $β$ that are suddenly quenched to another heat reservoir with inverse temperature $β' = (1-γ)β$. Furthermore, we show the connection with multifractal thermodynamics. For the non-equilibrium case, we show that the dynamics of systems described by exponential Kolmogorov-Nagumo averages still observe a second law of thermodynamics and the H-theorem. We further discuss the applications of stochastic thermodynamics in those systems -- namely, the validity of fluctuation theorems -- and the connection with thermodynamic length. namic length.

cond-mat.stat-mech↗

arXiv2024

Null models for comparing information decomposition across complex systems

A key feature of information theory is its universality, as it can be applied to study a broad variety of complex systems. However, many information-theoretic measures can vary significantly even across systems with similar properties, making normalisation techniques essential for allowing meaningful comparisons across datasets. Inspired by the framework of Partial Information Decomposition (PID), here we introduce Null Models for Information Theory (NuMIT), a null model-based non-linear normalisation procedure which improves upon standard entropy-based normalisation approaches and overcomes their limitations. We provide practical implementations of the technique for systems with different statistics, and showcase the method on synthetic models and on human neuroimaging data. Our results demonstrate that NuMIT provides a robust and reliable tool to characterise complex systems of interest, allowing cross-dataset comparisons and providing a meaningful significance test for PID analyses.

cs.IT↗

arXiv2024

Characterising high-order interdependence via entropic conjugation

High-order phenomena play crucial roles in many systems of interest, but their analysis is often highly nontrivial. There is a rich literature providing a number of alternative information-theoretic quantities capturing high-order phenomena, but their interpretation and relationship with each other is not well understood. The lack of principles unifying these quantities obscures the choice of tools for enabling specific type of analyses. Here we show how an entropic conjugation provides a theoretically grounded principle to investigate the space of possible high-order quantities, clarifying the nature of the existent metrics while revealing gaps in the literature. This leads to identify novel notions of symmetry and skew-symmetry as key properties for guaranteeing a balanced account of high-order interdependencies and enabling broadly applicable analyses across physical systems.

cs.IT↗