Expectation-Biased Stochastic Resonance: Two Regimes of Timescale-Dependent Dimensionality in Biological Systems

Expectation-Biased Stochastic Resonance: Two Regimes of Timescale-Dependent Dimensionality in Biological Systems Ian Todd Sydney Medical School The University of Sydney Sydney, NSW, Australia ito[email protected].edu.au Abstract We distinguish two mechanisms by which effective dimensionality depends on timescale. Regime 1 (Measurement): An observer with bandwidth Band time τcan resolve ∼2Bτ modes (Slepian-Landau-Pollak)—this applies to digital sampling and recordings. Regime 2 (Coupling): Biological oscillators continuously couple through physical interactions. A 1 Hz circadian oscillator encodes environmental dynamics at fine temporal resolution (bounded by coupling bandwidth and phase noise)—information is set by coupling bandwidth and SNR, not carrier frequency. Effective dimensionality depends on coupling propagation time and topology, not time-bandwidth products. For neural systems, the constraint is establishing phase coherence through anatomical connections (τcouple ∼ms to s). For circadian rhythms, seasonal entrainment uses continuous coupling to photoperiod and temperature. Biological information processing exploits analog coupling consistent with channel capacity while exceeding discrete sampling constraints. We formalize both regimes and show most biological timescale-dependence reflects coupling dynamics. This explains why organisms achieve information densities appearing to violate sampling intuitions and connects to sub-Landauer limits where analog coupling operates near thermal noise. Our claims distinguish sampling-limited measurement from analog coupling, not contradict Shannon capacity. Keywords: expectation-biased stochastic resonance, working memory, magical number seven, predictive coding, neural oscillations, top-down control, top-down predictions, prefrontal cortex, structured noise, phase dynamics, analog coupling, timescales, consciousness, pareidolia, subLandauer limits 1 Introduction: The Analog-Digital Distinction 1.1 The Working Memory Puzzle A central question in cognitive neuroscience: why can we hold only 7±2 items in working memory [28]? With ∼1011 neurons in the human brain, this severe capacity limit seems paradoxical. Modern theories propose:5 •Discrete ”slots” with fixed capacity [24] •Resource limitations constraining precision [3] 1 •Neural noise limiting fidelity [26] We propose a fundamentally different explanation: working memory capacity reflects the number of stable resonant modes that can coexist in a high-dimensional coupled oscillator system. Prefrontal10 cortex (PFC) maintains multiple items by establishing distinct phase-locked configurations—each item corresponds to a stable attractor in the collective phase space [30]. The capacity in the 4–7 item range (estimates vary across paradigms [10]) emerges from the physics of coupled oscillators: beyond ∼7-9 stable modes, interference between attractors causes collapse into lower-dimensional configurations.15 This is analogous to resonant modes in physical systems (vibrating membranes, electromagnetic cavities): only certain discrete configurations can exist stably. Working memory doesn’t store items in ”slots”—it maintains them as distinct resonant patterns in neural phase space. When you try to add an 8th or 9th item, the phase space becomes overcrowded and existing patterns interfere, causing degradation or loss.20 1.2 The Central Confusion How many degrees of freedom participate in a biological process? Standard information theory suggests: if a system oscillates at frequency fand you observe for time τ, you can extract ∼2fτ bits of information (Nyquist-Shannon). This implies that slow oscillators carry less information than fast ones.25 But this is misleading when applied to analog systems. A circadian clock oscillating at 1 cycle per 24 hours does not carry only ”1/86400 Hz worth” of information. It continuously tracks: •Light intensity variations (millisecond timescale) •Temperature fluctuations (second to hour timescale) •Metabolic state (minute to hour timescale)30 •Social zeitgebers (minute to hour timescale) All this information is encoded in the continuous phase trajectory ϕ(t) of the oscillator. The 24h period is merely the carrier frequency. The actual information content scales with the coupling bandwidth, not the oscillator frequency. Key insight: Digital measurement obeys Shannon limits. Analog coupling does not obey35 sampling-theorem constraints—and biology implements this through expectation-biased stochastic resonance (EBSR): actively generating structured ”noise” via internal models that match expected signal structure, creating matched filters that sympathetically lock to weak signals while producing meaningful false positives (seeing tigers in leaves when anxious). EBSR increases the effective dimensionality of detection space by shaping the ”noise bucket” to align with expected signal40 manifolds, enabling high-dimensional computation in unmeasurable regimes [38]. This is our central novel contribution. 1.3 Two Regimes of Dimensionality We formalize two distinct mechanisms: 2 Regime 1: Measurement-Limited Dimensionality (Digital) Context: External observer with digital recording, finite bandwidth B(Hz), integration time τ Bound: Dmeas eff (τ)≲2Bτ (Slepian-Landau-Pollak) Mechanism: Nyquist sampling limits number of resolvable orthogonal modes Examples: EEG recording, gravitational wave detection, discrete computational models Note: Bis measured in Hz throughout; the factor of 2 arises from positive and negative frequencies. 45 Regime 2: Coupling-Limited Dimensionality (Analog) Context: Intrinsic dynamics of physically coupled oscillators Bound: Dcouple eff (τ)∼Ncoupled(τ) (depends on coupling topology and propagation time, NOT on oscillator frequency) Mechanism: Coupling requires time to propagate and establish phase coherence. Effective dimensionality = number of oscillators that have established coherent coupling within time τ. Examples: Neural assemblies phase-locking through synapses, circadian clocks entraining to environment, metabolic networks coordinating through diffusion Critical distinction: In Regime 2, a 1 Hz oscillator can encode high-frequency information through phase modulation ϕ(t)=ω0t+δϕ(t), where δϕ(t) tracks environmental variations up to the coupling bandwidth and phase-noise floor. This is how radio FM works—the carrier frequency doesn’t limit information content.50 1.4 Biological Relevance Most biological information processing operates in Regime 2. Organisms don’t ”measure” their environment by taking discrete samples at the oscillator frequency. They continuously couple through: •Synaptic transmission (chemical diffusion, receptor dynamics)55 •Metabolic coupling (enzyme activity, substrate availability) •Hormonal signaling (secretion, transport, receptor binding) •Transcriptional coupling (TF binding, mRNA production, protein translation) All these are analog processes operating continuously in time. The effective dimensionality depends on how many coupled subsystems have established phase coherence, not on how many60 ”samples” have been taken. Regime 1 (measurement) is relevant when we analyze biological systems with digital tools (computers, discrete recordings). The Shannon bound limits what we can measure, not what the organism is doing. 1.5 Structure65 Section 2 formalizes measurement-limited (digital, Shannon-bound) dimensionality and establishes when sampling constraints apply. Section 3 develops coupling-limited (analog, EBSR-enhanced) 3 regimes and introduces expectation-biased stochastic resonance as a novel mechanism. Section 4 applies EBSR to prefrontal control, working memory capacity, and neural timescales. Section 5 addresses consciousness and predictive coding integration. Section 6 discusses evolutionary ad-70 vantages. Section 7 provides testable predictions distinguishing EBSR from standard SR, with emphasis on false positive structure and working memory phenomena. Section 8 addresses clinical implications and open questions. 2 Regime 1: Measurement-Limited Dimensionality 2.1 Measurement Protocol75 An external observer is defined as a measurement protocol M(τ, B, ε): •Integration time τ(observation window) •Effective bandwidth B(Hz, frequency range of measurement apparatus) •Resolution threshold ε(minimum distinguishable signal) This applies to: digital recording systems, computational models with discrete timesteps, ex-80 perimental setups with finite sampling rates. 2.2 Slepian-Landau-Pollak Bound For a windowed measurement, define projection operators: •Tτ: time-limit to window [0, τ] •BB: band-limit to frequencies [−B,B]85 The Slepian-Landau-Pollak theorem [34, 20] characterizes: TτBBTτψ=λψ (1) The eigenfunctions are prolate spheroidal wave functions (PSWFs). The number of eigenvalues λi≥εis: rankε(TτBBTτ)≈2Bτ (2) where Bis measured in Hz (the factor of 2 accounts for positive and negative frequencies; for angular frequency use B/2π).90 This is the Shannon number—the number of orthogonal modes a digital measurement system can resolve [35, 21]. Key point: This is a measurement constraint, not a constraint on the underlying system. If the system dynamics have bandwidth Bsys > Bmeas, the observer misses information. 2.3 Participation Ratio for Measured Data95 For measured data with Nchannels, the windowed covariance: Cτ=E[xτx⊤ τ] (3) 4 Effective dimensionality via participation ratio: Dmeas eff (τ) = PN i=1 λi(τ)2 PN i=1 λi(τ)2≲2Bmeasτ(4) Example: EEG recorded at 1 kHz (Bmeas = 500 Hz) for 1 s yields Dmeas eff ≲1000 resolvable modes. But the underlying neural dynamics might have far more degrees of freedom coupling at higher bandwidths.100 3 Regime 2: Coupling-Limited Dimensionality (Analog) 3.1 The Mechanism: Stochastic Resonance and Tunable Noise Before formalizing continuous phase dynamics, we establish the mechanism by which biological systems exploit analog coupling: stochastic resonance (SR). 3.1.1 The Local Resonant Copy Principle105 Consider measuring a distant 440 Hz tone. Two approaches: Digital (Regime 1): Microphone samples at ≥880 Hz (Nyquist), ADC discretizes, FFT extracts the frequency. Limited by sampling rate, quantization noise, and computational resources. Analog (Regime 2): Place a 440 Hz tuning fork nearby. Sound waves continuously couple to the fork’s mechanical oscillation. The fork resonates sympathetically—building amplitude over110 cycles through continuous energy transfer. No sampling, no discretization. The fork acts as a passive bandpass filter, amplifying 440 Hz while rejecting noise. Why this works better: •Resonant amplification: The fork’s Q-factor provides gain without active electronics •Phase-locked detection: Continuous coupling preserves phase information115 •Energy efficiency: Passive mechanical resonance, no power consumption •Noise rejection: Frequency selectivity suppresses broadband noise This is standard physics (sympathetic vibration), but the principle extends to all biological oscillators: neurons, metabolic cycles, circadian clocks. 3.1.2 Stochastic Resonance: Noise-Enhanced Detection120 In real biological systems, signals are often sub-threshold—too weak to trigger response without assistance. Stochastic resonance exploits added noise to push weak signals over detection thresholds [43, 27]. Mechanism: A weak periodic signal s(t) (e.g., distant 440 Hz tone, faint visual edge, subtle metabolic perturbation) is below detection threshold. Adding moderate noise ξ(t) causes the system125 to occasionally cross threshold in phase with the signal. Averaging over time reveals the signal that was previously invisible. Classic examples [31, 16, 4, 8]: •Crayfish mechanoreceptors detecting water vibrations 5 •Paddlefish electroreception for prey130 •Human tactile sensing enhanced by vibration •Auditory neurons detecting faint tones (10–20 dB improvement) Critical insight: SR is not limited to sensory systems. It operates throughout biology— metabolic oscillations, circadian clocks, neural synchronization, immune responses. It is a biological universal.135 3.1.3 Novel Framework: Expectation-Biased Stochastic Resonance (EBSR) Standard stochastic resonance uses unstructured white noise to enhance detection. We propose a fundamentally different mechanism operating in biological systems: expectation-biased stochastic resonance (EBSR). The key insight: The ”noise” is not random—it is structured fluctuation generated by an140 internal model with the same dimensionality and structure as the expected signal. The bucket analogy: Imagine trying to capture signals arriving from a high-dimensional space. Noise in stochastic resonance increases the dimensionality of your detection space—it expands your ”bucket.” Standard SR adds noise uniformly, expanding the bucket as a hypersphere (radius r) in all Dtotal dimensions equally, with volume145 Vwhite =πDtotal/2 Γ(Dtotal/2 + 1)rDtotal .(5) EBSR shapes the bucket to match the expected signal structure: you expand only in the Dsignal dimensions where the signal lives, creating an elongated hyperellipsoid aligned with the expected manifold, with volume VEBSR =πDsignal/2 Γ(Dsignal/2 + 1) Dsignal Y i=1 ri,(6) where ri≫ralong signal dimensions and ri≈0 elsewhere. For the same detection probability (captured volume in signal subspace), EBSR concentrates resources strategically, dramatically150 increasing detection efficiency while maintaining specificity. Mechanism: Suppose you’re searching for a signal produced by a 15-dimensional system (e.g., a tiger: visual stripes, motion patterns, spatial context, temporal dynamics, etc.). Rather than adding generic white noise, the organism generates ”noise” by running an internal 15D model that produces tiger-like fluctuations:155 xperceived(t)=xsensory(t)+α·xmodel(t) (7) where: •xsensory(t)∈RDtotal is the actual sensory input (potentially weak/ambiguous) •xmodel(t)∈RDtotal is internally-generated ”noise” from a 15D tiger model (living in Dsignal = 15 dimensional subspace) •αis the gain (modulated by expectation, anxiety, priors)160 6 Note: We assume Gaussian noise for tractability and linear addition as a first approximation; extensions to nonlinear coupling and non-Gaussian statistics are straightforward. When a real tiger is present: The sensory input xsensory has the same 15D structure as xmodel. They constructively interfere, producing strong detection. This is matched filtering—the internal model acts as a resonator tuned to the expected signal. The shaped ”bucket” aligns165 perfectly with the incoming signal direction in high-dimensional space. When no tiger is present: The sensory input (leaves, shadows) has different structure. But with high α(high anxiety), the internal ”noise” occasionally reaches threshold, producing false positives: seeing tigers in the leaves. The bucket is so elongated in tiger-space that random fluctuations occasionally land inside it.170 Why this is not standard SR: •Standard SR: Noise is white/unstructured, expands bucket spherically in all dimensions equally •EBSR: ”Noise” is structured, expands bucket only along expected signal manifold (elongated ellipsoid)175 •Standard SR: Cannot explain false positives with specific content (why tigers, not random blobs?) •EBSR: False positives have the structure of the internal model (pareidolia) Dimensional expansion through EBSR: By generating structured noise, the system effectively increases the dimensionality of its detection space from the ambient Dtotal to a focused180 subspace Dsignal. The ”bucket” becomes a high-dimensional needle oriented toward expected signals, maximizing sensitivity while maintaining specificity through dimensional selectivity [38]. Sensitivity and specificity tradeoff: EBSR addresses the fundamental diagnostic challenge taught in medical training: balancing sensitivity (detecting true positives) against specificity (avoiding false positives). By generating noise only in the expected signal subspace, EBSR achieves:185 •High sensitivity: Strong amplification of signals matching the internal model (constructive interference in the Dsignal subspace) •Maintained specificity: Rejection of signals in irrelevant dimensions (no noise amplification in the remaining Dtotal −Dsignal dimensions) For a 15D tiger signal embedded in ∼106D visual space, EBSR explores 15 dimensions; white190 noise explores all 106. This yields an order-of-magnitude specificity advantage that can scale like Dtotal/Dsignal under matched-template detection (heuristic, depending on feature geometry and detection criterion): Specificity gain ∼Dtotal Dsignal ≈106 15 ≈7×104-fold.(8) This is not about ”saving energy” per se—it’s about achieving detection selectivity by targeting the right feature dimensions, exactly analogous to clinical diagnostic tests with high sensitivity and195 specificity. Detection with matched structure. To formalize EBSR’s advantage, we use signal detection theory. Let m(t) be the internal-model template and y(t) = s(t)+n(t)+αm(t) the observed variable under EBSR. Using the matched statistic T=Ry(t)m(t)dt, detection performance is quantified by200 d′=E[T|H1]−E[T|H0] std[T|H0].(9) 7 As in classic stochastic resonance, d′(α) is unimodal with an optimal α; beyond this optimum, false alarms dominate. When maligns with s(matched structure), d′increases with αup to this optimum (classic SR peak), while false positives inherit the content of m—a hallmark prediction distinguishing EBSR from white-noise SR. Standard SR improves detectability (hit rate, d′), not raw linear SNR in the usual sense; EBSR extends this by providing structure-specific amplification.205 Evolutionary logic: •False positives (seeing phantom tigers): brief cost, avoid predator •False negatives (missing real tiger): death •Natural selection favors EBSR with high αin dangerous contexts The sensitivity-specificity balance is evolutionarily optimized: high sensitivity in threat detec-210 tion (better to flee from rustling leaves than miss a predator), with specificity maintained through dimensional targeting (don’t confuse tigers with birds). Neural implementation: Top-down projections from cortex to sensory areas carry predictions [15, 7]. These aren’t ”signals”—they’re fluctuating patterns generated by internal models. In predictive coding, prediction errors drive perception. But the predictions themselves can act as215 EBSR noise when weak sensory evidence requires augmentation. This connection between SR and predictive coding has been noted [41, 5], but the structured nature of the noise (matching expected signal dimensionality) is our novel contribution. Neuromodulator control: Anxiety, attention, arousal regulate α: •High anxiety (threat context): increase α→more false positives, fewer misses (high sensitiv-220 ity) •Low arousal (safe context): decrease α→fewer false positives, maintained specificity •Attention (task-specific): select which internal model generates noise [6] This explains: •Pareidolia (seeing faces in clouds [23], Jesus in toast): high αfor face detection model225 •Auditory verbal hallucinations: internal speech model generates EBSR ”noise” [9] •Anxiety-induced false alarms: threat models produce high-αEBSR •Perceptual learning: refining internal models improves EBSR matching Connection to sub-Landauer limits: At signal energies Es∼kBT, discrete measurement fails. EBSR provides a solution: the internal model ”fills in” missing information through structured230 noise that resonates with weak coupling signals. By generating fluctuations in the signal subspace ξmodel(t) with amplitude σmodel, the system effectively reduces the timing uncertainty ∆trequired for detection, allowing biological information processing at thermal noise levels without requiring strong signals [37]. Novel prediction: The structure of false positives reveals the structure of internal models.235 Anxiety about specific threats (snakes vs. tigers) should produce false positives with corresponding specific structures. This is testable and distinguishes EBSR from standard SR. 8 Table 1: Comparison of standard stochastic resonance vs. expectation-biased stochastic resonance Feature Standard SR EBSR (Novel) Noise structure Unstructured white noise, uniform across all dimensions Structured noise from internal model, concentrated in signal subspace Dimensional expansion Expands detection ”bucket” spherically (hypersphere) in all dimensions equally Expands bucket as elongated hyperellipsoid aligned with expected signal manifold Amplification Enhances all signals equally Preferentially amplifies expected signal structure (matched filtering) False positives Non-specific, random Specific, meaningful content matching internal model (pareidolia) Dimensional exploration Explores all Dtotal dimensions Explores only Dsignal dimensions Tunability Fixed noise level or simple amplitude modulation Structured noise adapts to task demands and priors via neuromodulators Detection metric Improves d′via noise Improves d′via matched template correlation Examples Sensory thresholds in simple detection Working memory maintenance, threat detection, predictive perception 3.1.4 SR and Analog Coupling: The Connection Stochastic resonance is the implementation of analog coupling in noisy, near-thermal environments: •Regime 2 requires weak-signal detection: Biological oscillators couple through diffusion,240 synaptic transmission, hormonal signaling—all operating near thermal noise (∼kBT). •SR enables detection at these levels: Noise pushes sub-threshold signals into the detectable range without requiring higher signal energy. •Phase-locking via noise: SR doesn’t just detect presence—it enables phase coherence. Noise helps oscillators lock to weak coupling signals, establishing the coherent dynamics that245 define Dcouple eff . Lock-in amplifier analogy: Professional instruments use local reference oscillators (the ”tuning fork”) multiplied with the input signal, then low-pass filtered. This demodulates the signal, shifting the resonant component to DC while averaging noise. Phase-sensitive detection achieves sensitivities down to nanovolts—far below digital FFT methods. Biology does this organically250 through SR-enhanced coupling. Connection to sub-Landauer limits: SR allows information processing at signal energies Es∼kBTwhere discrete measurement fails [37]. Continuous analog coupling with noise-enhanced detection bypasses the timing-energy constraints that limit digital processing. 9 limit of 500 Hz on resolvable frequencies. The neural system may operate at higher bandwidths465 via analog coupling + EBSR, but our measurements are Shannon-limited. Computational models: Discrete-timestep simulations inherently operate in Regime 1. To capture EBSR dynamics, models need: (i) continuous-time integration, (ii) structured noise terms ξmodel(t) from internal predictive models, (iii) neuromodulatory control of αgain. Most current models lack these features and inadvertently impose sampling constraints biology circumvents.470 Implication: When neural data shows apparent dimensionality limits (e.g., from PCA on spike trains), this may reflect measurement constraints, not biological limits. The EBSR-enhanced analog dynamics might operate in higher dimensions inaccessible to digital recording. 8 Testable Predictions 8.1 Distinguishing Regime 1 vs. Regime 2475 Prediction 1: Information Density Beyond Sampling Constraints Test: Measure mutual information between circadian clock phase and environmental variables at different timescales. Regime 1 prediction: I≲2Bclockτwhere Bclock ∼1/(24 h) ∼10−5Hz. For τ= 1 day, I≲1 bit. Regime 2 prediction: Iscales as O(Bcoupleτ) at fixed SNR, where Bcouple ∼mHz (photoperiod precision). For τ= 1 day, I∼103bits (SNR permitting). Method: Measure phase shifts ∆ϕin response to brief environmental perturbations (light pulses, temperature steps). Precision of ∆ϕestimates Bcouple. Prediction 2 (Neural phase tracking with SR): Single neurons continuously track synaptic inputs at kHz bandwidths despite participating in slow (Hz) oscillations. Use intracellular recording to measure membrane potential V(t) during theta oscillations (4–8 Hz). The power spectrum of V(t) should extend to kHz (individual EPSPs), far exceeding the 4–8 Hz ”carrier frequency.”480 Additionally: Add moderate noise (current injection or pharmacological) and measure detection threshold for weak synaptic inputs. Should find optimal noise level (SR signature) that minimizes threshold—confirming noise-enhanced analog coupling. Prediction 3 (Metabolic coupling): Glycolytic oscillations (period ∼min) continuously respond to glucose perturbations on second timescales. Perturb glucose at t= 0 during glycolytic485 oscillation; measure phase shift ∆ϕ(t). Should see immediate response (∆t≪Tosc), confirming analog coupling. Prediction 4 (Consciousness and coupling bandwidth): Conscious states should show higher neural coupling bandwidth than unconscious states, independent of primary oscillation frequencies. Measure effective coupling bandwidth from EEG phase-locking during waking vs. deep490 sleep. Waking should show broadband coupling (Hz to tens of Hz); sleep shows narrow-band coupling. Prediction 5 (Anesthesia disrupts coupling): General anesthetics reduce consciousness by disrupting anatomical coupling (synaptic transmission), not by changing oscillation frequencies. Measure τcouple from cross-correlation timescales during waking vs. anesthesia. Anesthesia should495 increase τcouple (weaker coupling), reducing Deff even if oscillation frequencies unchanged. 16 Prediction 6: Expectation-Biased Stochastic Resonance (EBSR) Core prediction: Internal models generate structured ”noise” that preferentially amplifies expected signals and produces structured false positives. Test 1 - False positive structure: Prime subjects with specific threats (snakes/tigers/spiders), present ambiguous visual noise, measure false alarm content. EBSR: False positives match primed threat. Standard SR: Non-specific. Test 2 - Dimensionality matching: Train subjects on signals with varying dimensionality (1D tone vs. 15D scene), record pre-stimulus neural activity (fMRI/MEG), compute participation ratio. EBSR: Pre-stimulus noise dimensionality matches expected signal. Standard SR: High-D/unstructured. Test 3 - Anxiety modulation: Induce threat-specific anxiety (snake videos), present ambiguous stimuli with varying signal strength. EBSR: Anxiety increases snake-specific false positives and sensitivity for snake-like patterns. Standard SR: All false positives increase equally. Test 4 - Neural decoding: Use decoded neurofeedback to identify object-specific patterns (tigers, faces), measure spontaneous reactivation pre-stimulus, correlate with detection and false alarm content. EBSR: Spontaneous reactivation predicts both detection and false alarm structure. Key distinction: Standard SR cannot explain why false positives have specific meaningful content. EBSR predicts false positive structure reveals internal model content. 8.2 Practical Methods Estimating coupling bandwidth: For oscillator with phase ϕ(t): 1. Apply brief perturbation at t= 0500 2. Measure phase response curve: ∆ϕ(∆t) for different perturbation timings ∆twithin cycle 3. Bandwidth: Bcouple ∼1/∆tmin where ∆tmin is finest resolvable timing Estimating coupling propagation time: For coupled oscillator network: 1. Perturb oscillator iat t= 0 2. Measure time to phase-locking with oscillator j:τij 505 3. Coupling time: τcouple ∼τij/dij where dij is network distance 9 Discussion 9.1 The Fundamental Distinction for Neuroscience The key conceptual advance: distinguishing measurement from intrinsic neural dynamics. Regime 1 (Measurement): Shannon bound applies because digital recording/computation510 discretizes time. The 2Bτ limit constrains what we can measure from neural systems, not what neurons are doing. Regime 2 (EBSR-Enhanced Coupling): Neural networks are analog continuous dynamical systems with structured noise from PFC. Information is encoded in continuous phase trajectories, 17 with top-down predictions acting as matched filters. Sampling constraints don’t apply to the515 biology. Neural information processing operates in Regime 2. Synaptic transmission, membrane potential dynamics, neuromodulation—all are analog processes operating continuously. The apparent ”discrete” events (action potentials) are markers in continuous dynamics, not fundamental discretization. PFC-generated EBSR noise provides the ”extra” dimensionality that discrete520 feedforward models lack. 9.2 Implications for Computational Neuroscience Standard models use discrete-time (Regime 1) to simulate Regime 2 systems. Missing ingredients: 1. Structured noise from internal models: Most models treat noise as nuisance to minimize. But biology actively generates structured noise (ξmodel) for signal enhancement. Models525 incorporating EBSR might better capture: •Working memory maintenance without recurrent excitation loops •Attention effects without explicit gain modulation •False alarms with meaningful content •The ”warmth” of perception (continuous probabilistic inference, not discrete decisions)530 2. Top-down predictions as active process: Predictive coding models [15] treat predictions as signals. EBSR reveals they’re also noise generators—fluctuating activity patterns that lower thresholds. This dual role (signal + noise) may explain why PFC lesions impair both prediction generation AND perceptual robustness. 3. Neuromodulator control of α:Dopamine, acetylcholine, norepinephrine regulate EBSR535 gain. Models should include α(t) as a state variable modulated by task demands, arousal, and learning. This might explain: •Why stimulants improve attention (increased α) •Why anxiolytics reduce false alarms (decreased α) •Why schizophrenia shows hallucinations (unregulated α)540 Connection to high-dimensional coherence: EBSR is the mechanism by which biological systems maintain and exploit high-dimensional coherence in thermodynamically constrained regimes [38]. By generating structured noise that expands detection space along expected signal manifolds, organisms achieve computational capacities that would be inaccessible through discrete enumeration. The ”bucket shaping” is not metaphor but mechanism: internal models literally545 define the geometry of detection space in high dimensions, enabling intelligence to operate at the boundaries of physical measurability. Recommendation for modelers: Use continuous-time differential equations with explicit EBSR terms (ξwhite +α·ξmodel) rather than discrete updates. Include PFC modules that generate structured noise matching task-relevant features. Allow neuromodulators to tune αbased on550 context. 18 9.3 Open Questions 1. [Priority] Can we rigorously quantify information capacity of analog coupling + SR vs. sampling-limited digital? Under what conditions does noise-enhanced analog provide advantage?555 2. What determines Bcouple in different biological systems? Is it limited by molecular diffusion, membrane time constants, SR dynamics, or something else? 3. How do organisms regulate both τcouple and SR noise levels (e.g., via neuromodulators changing synaptic strength and stochastic firing)? Are these independent control parameters? 4. Can artificial systems exploit analog coupling + SR to circumvent sampling constraints?560 (E.g., neuromorphic hardware with continuous-time dynamics and tunable noise.) 5. What is the fundamental limit on analog coupling precision? Is there a noise floor set by thermodynamics (∼kBT)? Does SR saturate at some optimal noise level? 6. Can SR be ”learned” or adapted? Do organisms adjust optimal noise parameters through experience (perceptual learning), or are they genetically determined? This ties to the broader565 question of how internal models are refined. 9.4 Clinical Implications Understanding EBSR opens new therapeutic avenues: Schizophrenia: Hallucinations may reflect excessive EBSR gain (αtoo high), where uncontrolled internal models generate structured noise that reaches perception threshold without external570 evidence. Antipsychotics may work partly by reducing α. PTSD: Hypervigilance reflects overactive threat models with high α, producing false alarms specific to traumatic content. Cognitive behavioral therapy may recalibrate internal models, reducing their contribution to EBSR noise and lowering αfor threat-related features. Anxiety disorders: Generalized anxiety may involve chronically elevated αacross multiple575 threat models, increasing false positive rates broadly. Anxiolytics (e.g., benzodiazepines) may work by reducing EBSR gain, trading sensitivity for specificity. Attention deficits: ADHD may involve insufficient αfor task-relevant features or inability to sustain structured noise over time, explaining distractibility and working memory deficits. Treatment strategies could target the balance between detection sensitivity and false positive580 rate by modulating: •Internal model content (cognitive therapy, exposure) •EBSR gain α(pharmacology, neuromodulation) •Prefrontal top-down control (cognitive training, neurofeedback) 10 Conclusion585 We have introduced expectation-biased stochastic resonance (EBSR)—a novel mechanism explaining how prefrontal cortex maintains working memory, generates perception, and produces meaningful false positives. Unlike standard SR (unstructured noise), EBSR generates structured 19 ”noise” via internal models matching expected signal dimensionality. The ”bucket analogy” captures the essence: EBSR shapes the detection space to align with expected signals, expanding590 dimensionality strategically rather than uniformly. Key cognitive neuroscience implications: 1. Working memory capacity: The 4–7 item capacity range emerges from constraints on maintaining EBSR noise for multiple items (∼5 items ×10-15 features each), not from fundamental representational limits. PFC generates structured activity (αgain) that maintains weak traces595 through resonant amplification, providing a mechanistic complement to both slot and resource theories [30, 25]. 2. Attention as EBSR gain control: Top-down attention from PFC [6] increases αfor attended features, generating stronger EBSR noise that enhances detection while producing featurespecific false alarms (Posner cueing effects).600 3. Consciousness requires structured noise: Global coherence across distributed assemblies (consciousness timescale ∼100 ms to 3 s) is maintained by PFC-generated EBSR noise propagating via beta oscillations [2]. Without sustained structured noise, representations fragment (anesthesia, distraction). 4. False positives reveal internal models: Pareidolia (faces in clouds [23]), hallucinations605 [9], and anxiety-induced false alarms all reflect high-αEBSR from overactive internal models. The content of false positives diagnoses which models are generating excessive noise. Two regimes distinguished: Regime 1 (Measurement): Digital recording systems obey Shannon bound Deff ≲2Bτ. Applies to experimental data acquisition and discrete computational models.610 Regime 2 (EBSR-Enhanced Coupling): Biological neural networks use continuous analog dynamics with EBSR. PFC generates structured noise only in expected signal subspace (15D for tiger vs. 106D visual cortex), providing ∼104-fold specificity advantage through dimensional selectivity. Information capacity is set by coupling bandwidth and SNR, not carrier frequency, consistent with information-theoretic bounds while circumventing sampling constraints.615 Novel predictions distinguishing EBSR from standard SR: •False positives have specific structure matching primed threats (snakes vs. tigers) •Pre-stimulus spontaneous activity dimensionality matches expected signal dimensionality •Anxiety increases threat-specific false alarms, not generic false alarms •Decoded neural patterns predict both detection AND false alarm content620 This framework unifies stochastic resonance, predictive coding [15, 7], and working memory through one mechanism: structured expectation-driven noise generation. Prefrontal cortex doesn’t just represent—it actively generates probabilistic templates that shape perception through resonant amplification of matching evidence. Broader implications: EBSR resolves why slow neural oscillations encode fine-grained infor-625 mation (analog coupling + structured noise, not discrete sampling), explains psychiatric conditions (mistuned internal models), and shows why evolution favored analog over digital processing (specificity through dimensionality-targeted noise). Clinical relevance: Hallucinations in schizophrenia may reflect excessive EBSR gain (αtoo high); PTSD hypervigilance reflects overactive threat models. Therapeutic interventions (cognitive630 behavioral therapy recalibrating models, pharmacology modulating α) may work by rebalancing the sensitivity-specificity tradeoff. Understanding EBSR could guide treatment strategies targeting internal model content and gain parameters. 20 The ”noise” in neural information processing is not random—it’s structured prediction, and its content encodes expectation.635 Citation Diversity Statement We acknowledge the importance of diverse citation practices. This manuscript cites works from a range of research groups across cognitive neuroscience, systems neuroscience, theoretical neuroscience, and biophysics. Author Contributions640 Ian Todd: Conceptualization, Formal Analysis, Writing - Original Draft, Writing - Review & Editing. Use of AI in Manuscript Preparation AI language models were used in the preparation of this manuscript. Claude (Anthropic) was used for drafting. GPT (OpenAI) and Grok (xAI) were used for error-checking and proofreading. All645 scientific content, conceptual framework, and final editorial decisions remain the sole responsibility of the author. References [1] Allada, R., & Bass, J. (2021). Circadian mechanisms in medicine. New England Journal of Medicine 384(6), 550–561. https://doi.org/10.1056/NEJMra1802337650 [2] Bastos, A. M., Usrey, W. M., Adams, R. A., Mangun, G. R., Fries, P., & Friston, K. J. (2012). Canonical microcircuits for predictive coding. Neuron 76(4), 695–711. https://doi.org/10. 1016/j.neuron.2012.10.038 [3] Bays, P. M., & Husain, M. (2008). Dynamic shifts of limited working memory resources in human vision. Science 321(5890), 851–854. https://doi.org/10.1126/science.1158023655 [4] Benzi, R., Sutera, A., & Vulpiani, A. (1981). The mechanism of stochastic resonance in climatic change. Tellus 34(1), 10–16. https://doi.org/10.1111/j.2153-3490.1981.tb01731.x [5] Bialek, W. (2012). Biophysics: Searching for Principles. Princeton University Press. [6] Buschman, T. J., & Miller, E. K. (2007). Top-down versus bottom-up control of attention in the prefrontal and posterior parietal cortices. Science 315(5820), 1860–1862. https://doi.660 org/10.1126/science.1138071 [7] Clark, A. (2013). Whatever next? Predictive brains, situated agents, and the future of cognitive science. Behavioral and Brain Sciences 36(3), 181–204. https://doi.org/10.1017/ S0140525X12000477 [8] Collins, J. J., Chow, C. C., & Imhoff, T. T. (1996). Stochastic resonance without tuning.665 Nature 376(6537), 236–238. https://doi.org/10.1038/376236a0 21 [9] Corlett, P. R., Taylor, J. R., Wang, X.-J., Fletcher, P. C., & Krystal, J. H. (2011). Toward a neurobiology of delusions. Progress in Neurobiology 92(3), 345–369. https://doi.org/10. 1016/j.pneurobio.2010.06.007 [10] Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental670 storage capacity. Behavioral and Brain Sciences 24(1), 87–114. https://doi.org/10.1017/ S0140525X01003922 [11] Cutler, C., and Flanagan, ´ E. E. (1994). Gravitational waves from merging compact binaries: How accurately can one extract the binary’s parameters? Phys. Rev. D 49, 2658–2697. https: //doi.org/10.1103/PhysRevD.49.2658675 [12] Dehaene, S., & Changeux, J.-P. (2011). Experimental and theoretical approaches to conscious processing. Neuron 70(2), 200–227. https://doi.org/10.1016/j.neuron.2011.03.018 [13] Demir, A., Mehrotra, A., & Roychowdhury, J. (2000). Phase noise in oscillators: a unifying theory and numerical methods for characterization. IEEE Transactions on Circuits and Systems I 47(5), 655–674. https://doi.org/10.1109/81.847872680 [14] Faisal, A. A., Selen, L. P. J., & Wolpert, D. M. (2008). Noise in the nervous system. Nature Reviews Neuroscience 9(4), 292–303. https://doi.org/10.1038/nrn2258 [15] Friston, K. (2005). A theory of cortical responses. Philosophical Transactions of the Royal Society B 360(1456), 815–836. https://doi.org/10.1098/rstb.2005.1622 [16] Gammaitoni, L., H¨anggi, P., Jung, P., & Marchesoni, F. (1998). Stochastic resonance. Reviews685 of Modern Physics 70(1), 223–287. https://doi.org/10.1103/RevModPhys.70.223 [17] Goldbeter, A. (1996). Biochemical Oscillations and Cellular Rhythms. Cambridge University Press. https://doi.org/10.1017/CBO9780511608193 [18] H¨anggi, P. (2002). Stochastic resonance in biology: How noise can enhance detection of weak signals and help improve biological information processing. ChemPhysChem 3(3), 285–290.690 https://doi.org/10.1002/1439-7641(20020315)3:3<285::AID-CPHC285>3.0.CO;2-A [19] Johnson, C. H. (1999). Forty years of PRCs—what have we learned? Chronobiology International 16(6), 711–743. https://doi.org/10.3109/07420529909016940 [20] Landau, H. J., and Pollak, H. O. (1961). Prolate spheroidal wave functions, Fourier analysis and uncertainty—II. Bell System Technical Journal 40(1), 65–84. https://doi.org/10.1002/695 j.1538-7305.1961.tb03977.x [21] Landau, H. J., and Widom, H. (1980). The eigenvalue distribution of time and frequency limiting. J. Math. Anal. Appl. 77(2), 469–481. https://doi.org/10.1016/0022-247X(80) 90241-3 [22] Lisman, J. E., & Idiart, M. A. P. (1995). Storage of 7 ±2 short-term memories in oscillatory700 subcycles. Science 267(5203), 1512–1515. https://doi.org/10.1126/science.7878473 [23] Liu, J., Li, J., Feng, L., Li, L., Tian, J., & Lee, K. (2014). Seeing Jesus in toast: Neural and behavioral correlates of face pareidolia. Cortex 53, 60–77. https://doi.org/10.1016/j. cortex.2014.01.013 22 [24] Luck, S. J., & Vogel, E. K. (2013). Visual working memory capacity: from psychophysics and705 neurobiology to individual differences. Trends in Cognitive Sciences 17(8), 391–400. https: //doi.org/10.1016/j.tics.2013.06.006 [25] Lundqvist, M., Rose, J., Herman, P., Brincat, S. L., Buschman, T. J., & Miller, E. K. (2016). Gamma and beta bursts underlie working memory. Neuron 90(1), 152–164. https://doi.org/ 10.1016/j.neuron.2016.02.028710 [26] Ma, W. J., Husain, M., & Bays, P. M. (2014). Changing concepts of working memory. Nature Neuroscience 17(3), 347–356. https://doi.org/10.1038/nn.3655 [27] McDonnell, M. D., & Abbott, D. (2009). What is stochastic resonance? Definitions, misconceptions, debates, and its relevance to biology. PLoS Computational Biology 5(5), e1000348. https://doi.org/10.1371/journal.pcbi.1000348715 [28] Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review 63(2), 81–97. https://doi.org/ 10.1037/h0043158 [29] Miller, E. K., & Cohen, J. D. (2001). An integrative theory of prefrontal cortex function. Annual Review of Neuroscience 24, 167–202. https://doi.org/10.1146/annurev.neuro.24.1.167720 [30] Miller, E. K., Lundqvist, M., & Bastos, A. M. (2018). Working memory 2.0. Neuron 100(2), 463–475. https://doi.org/10.1016/j.neuron.2018.09.023 [31] Moss, F., & Wiesenfeld, K. (1994). The benefits of background noise. Scientific American 273(2), 66–69. https://doi.org/10.1038/scientificamerican0895-66 [32] Pikovsky, A., Rosenblum, M., & Kurths, J. (2001). Synchronization: A Universal725 Concept in Nonlinear Sciences. Cambridge University Press. https://doi.org/10.1017/ CBO9780511755743 [33] P¨oppel, E. (2009). Pre-semantically defined temporal windows for cognitive processing. Philosophical Transactions of the Royal Society B 364(1525), 1887–1896. https://doi.org/10. 1098/rstb.2009.0015730 [34] Slepian, D., and Pollak, H. O. (1961). Prolate spheroidal wave functions, Fourier analysis and uncertainty—I. Bell System Technical Journal 40(1), 43–63. https://doi.org/10.1002/j. 1538-7305.1961.tb03976.x [35] Slepian, D. (1964). Prolate spheroidal wave functions, Fourier analysis and uncertainty—IV. Bell System Technical Journal 43(6), 3009–3057. https://doi.org/10.1002/j.1538-7305.735 1964.tb01037.x [36] Strogatz, S. H. (2000). From Kuramoto to Crawford: exploring the onset of synchronization in populations of coupled oscillators. Physica D 143(1–4), 1–20. https://doi.org/10.1016/ S0167-2789(00)00094-4 [37] Todd, I. (2025). The limits of falsifiability: Dimensionality, measurement thresholds, and the740 sub-Landauer domain in biological systems. BioSystems (accepted, in press). Preprint: Zenodo. https://doi.org/10.5281/zenodo.17247628 23 [38] Todd, I. (2025). Intelligence as High-Dimensional Coherence: Energy Landscapes, Phase Space, and the Physical Basis of Computation. Manuscript under review at BioSystems. Preprint: https://ssrn.com/abstract=5573915745 [39] Tononi, G., Boly, M., Massimini, M., & Koch, C. (2016). Integrated information theory: from consciousness to its physical substrate. Nature Reviews Neuroscience 17(7), 450–461. https://doi.org/10.1038/nrn.2016.44 [40] Vallisneri, M. (2008). Use and abuse of the Fisher information matrix in the assessment of gravitational-wave parameter-estimation prospects. Phys. Rev. D 77, 042001. https://doi.750 org/10.1103/PhysRevD.77.042001 [41] Ward, L. M. (2002). Dynamical cognitive science. MIT Press. [42] Wellens, T., Shatokhin, V., & Buchleitner, A. (2008). Stochastic resonance. Reports on Progress in Physics 71(10), 106001. https://doi.org/10.1088/0034-4885/71/10/106001 [43] Wiesenfeld, K., & Moss, F. (1995). Stochastic resonance and the benefits of noise: from ice755 ages to crayfish and SQUIDs. Nature 373(6509), 33–36. https://doi.org/10.1038/373033a0 24