--- title: Inference-time Plasticity Explained url: https://www.emergentmind.com/topics/inference-time-plasticity type: topic --- # Inference-time Plasticity Explained Inference-time plasticity denotes regimes in which synaptic efficacies or other adaptive parameters continue to change during ongoing operation, so that learning-like state change is part of the inference process rather than being confined to an offline training stage. In the literature, the plastic variable may be the momentary synaptic efficacy \(W_i(t)\propto w U R\) induced by short-term synaptic plasticity in spiking Boltzmann-style samplers, a short-term component \(\boldsymbol{F}^{(k)}_t\) added to resting weights \(\boldsymbol{W}^{(k)}\), continuously evolving STDP weights in unsupervised spiking networks, readout weights updated millisecond-by-millisecond in a timing decoder, or fast weights \(w_l(t)\) added to static Transformer parameters \(\tilde W_l\) within a sequence [1709.08166, 2009.06808, 2407.10028, 2510.14382, 2510.21908]. The unifying feature is that the system’s effective input–output map is altered online by local or structured plastic dynamics; the main exclusions are settings in which plasticity is used only to construct a fixed inference circuit prior to evaluation [2512.23728]. ## 1. Conceptual scope and definitional boundaries The term covers several distinct but related operational regimes. In the strongest sense, synapses or fast weights are updated while samples, tokens, or sensorimotor streams are being processed, with no separation between “learning” and “using” the model. Short-term synaptic plasticity in spiking generative networks is an instance of this strict form: the long-term weight matrix remains fixed, but instantaneous synaptic efficacy changes online and transiently during sampling, thereby reshaping the current energy landscape [1709.08166]. Sequence-level fast weights in decoder-only Transformers are another direct instance: a plastic component \(w_l(t)\) is initialized at the start of each sequence and updated at each step, so adaptation occurs inside the forward pass [2510.21908]. A weaker, hybrid sense appears in models that retain an explicit train/test distinction within trials yet still rely on cumulative online plasticity across nonstationary operating regimes. In the reservoir spiking model for joint “what” and “when” prediction, the “when” decoder learns online within trials, the “what” decoder is consolidated offline at trial end, and cue-only evaluation disables learning during the test phase; nevertheless, the weights are carried forward across blocks, permitting adaptation after block switches without resetting the architecture [2510.14382]. This suggests that inference-time plasticity can be localized to one computational pathway rather than being a property of the entire model. The boundary with adjacent notions is important. “Spike-Timing-Dependent Plasticity for Bernoulli Message Passing” uses STDP to train spiking modules that later implement sum-product updates for Bernoulli messages, but after training the “training layer” is removed and the network is used as a fixed inference circuit; this is plasticity for constructing an inference machine, not plasticity as the inference process itself [2512.23728]. “A study on the plasticity of neural networks” uses the word plasticity in a different sense again: the ability of a pretrained network to remain capable of reaching a well-generalizing solution after further adaptation, with the central issue being loss of plasticity under pretraining rather than online synaptic change during inference [2106.00042]. A further extension appears in work that treats plasticity itself as the latent object to be inferred. The STEER framework separates fast within-session neural dynamics from slow session-to-session plastic change and formalizes the latter by a latent recurrence \(z_{k+1}=g_\theta(z_k,u_k)\), so that the plasticity rule becomes identifiable, interpretable, and predictive under new stimulation schedules [2603.00213]. This broadens the topic from online parameter updates to principled inference over the law governing adaptation. | Regime | Plastic locus | Operational role | |---|---|---| | Local short-term synaptic dynamics | \(W_i(t)\propto w U R\) | Local reshaping of the current attractor landscape during sampling | | Hybrid online/offline predictive coding | “when” readout weights | Online within-trial timing adaptation with offline “what” consolidation | | Sequence-level fast weights | \(W_l(t)=\tilde W_l+w_l(t)\) | Rapid task-specific adaptation inside a sequence | ## 2. Short-term synaptic dynamics in probabilistic inference A central line of work treats inference-time plasticity as transient modification of effective synaptic strength during probabilistic sampling. In “Spiking neurons with short-term synaptic plasticity form superior generative networks,” the reference probabilistic model is a Boltzmann machine with binary states \(z_k\in\{0,1\}\) and distribution \[ B(z \mid W,b)=Z^{-1}\exp[-\beta E(z)], \qquad E(z)=-\frac{1}{2} z^T W z - z^T b. \] The spiking implementation uses conductance-based leaky integrate-and-fire neurons in the high-conductance state, where strong balanced Poisson background input makes firing probability approximately logistic. The membrane potential obeys \[ m \frac{d u}{d t} = l (l-u) + syn, \] with synaptic input \[ syn(t) = \sum_s \sum_i W_i(t)\,(rev_i-u)\exp\!\left[-(t-t^s)/syn\right], \] and activation in the high-conductance state approximates \[ p(z_k=1) \approx \sigma(\alpha f_k - \beta). \] Within this mapping, short-term plasticity is modeled by the Tsodyks–Markram equations, with postsynaptic impact \(P \propto w\cdot U\cdot R\), \[ \frac{dR}{dt} = \frac{1-R}{rec} - U\,R\,\delta(t-t_s), \qquad \frac{dU}{dt} = -\frac{U}{fac} + U_0(1-U)\,\delta(t-t_s). \] The key claim is that the plastic variable is the instantaneous synaptic efficacy during processing, not the long-term learned weight matrix; STP therefore acts as a local, spike-triggered, transient inference-time mechanism [1709.08166]. Its functional role is formulated as a local alternative to tempering. In multimodal generative models trained on structured data, deep attractor basins and high energy barriers impair mixing. Simulated tempering, adaptive simulated tempering, and CAST address this by globally altering inverse temperature \(\beta_T\), but they require extra computation, global state changes, and many samples that are not valid at the target temperature. STP instead weakens the attractor currently being visited by affecting only the efferent connections of simultaneously active neurons. Rather than flattening the entire energy landscape, it locally reshapes the landscape around the current state, functioning as an activity-dependent self-tempering process [1709.08166]. The quantitative results support that interpretation. In a 10-neuron target distribution, the Kullback–Leibler divergence between sampled and target distributions was minimized for STP settings around \(rec\approx 15\) ms, close to the synaptic time constant \(syn=10\) ms, and static synapses were not optimal. On the hard bar-image dataset, the STP network could still mix whereas ordinary Gibbs sampling often remained stuck; in the easy case, both mixed, but Gibbs spent about \(100\) times longer in the same mode before switching. On MNIST, a 3-layer hierarchical network with \(784\) visible, \(600\) hidden, and \(10\) label units, using \(U_0=0.01\) and \(rec=280\) ms, showed faster growth of the indirect sampling likelihood, more diverse samples after \(10^3\) samples (about \(10\) s biological time), and similar long-run ISL to Gibbs after \(>10^5\) samples. Discriminative performance changed little: Gibbs reached \(93.4\%\) test accuracy and the STP spiking network \(93.2\%\). Under imbalanced digit frequencies, the spiking network generated a more even class distribution over \(16{,}000\) samples and more balanced ambiguous completions than Gibbs or AST, indicating faster access to minority modes in finite-time inference [1709.08166]. A second formalization links inference-time plasticity directly to Bayes-optimal prediction in dynamic environments. “Optimality of short-term synaptic plasticity in modelling certain dynamic environments” decomposes synaptic efficacy as \[ \boldsymbol{G}^{(k)}_t=\boldsymbol{W}^{(k)}+\boldsymbol{F}^{(k)}_t, \] where \(\boldsymbol{W}^{(k)}\) is a resting weight and \(\boldsymbol{F}^{(k)}_t\) is a dynamic short-term component updated online by \[ \Delta\boldsymbol{F}^{(k)}_t=\gamma \boldsymbol{X}_t Q^{(k)}_t, \qquad \frac{d \boldsymbol{F}^{(k)}_t}{dt}=-\lambda \boldsymbol{F}^{(k)}_t. \] The central claim is that for environments whose observations transform continuously but randomly over time, the Bayes-optimal predictor or inferencer is equivalent to short-term spike-timing-dependent plasticity. In this view, recent input pulls the synapse toward the current observation, after which the synapse relaxes back toward its resting state. On OMNIST, an SNN trained on static MNIST achieved \(88.49\% \pm 0.12\%\) on MNIST test data, dropped to \(61.10\% \pm 0.48\%\) on OMNIST without ST-STDP, and rose to \(87.90\% \pm 0.29\%\) on OMNIST with ST-STDP enabled during inference; the same setup outperformed MLP, CNN, RNN, and LSTM baselines reported for that benchmark [2009.06808]. ## 3. Predictive timing, probability, and hybrid online plasticity in spiking reservoirs Inference-time plasticity also appears as pathway-specific online readout learning. In “Joint encoding of ‘what’ and ‘when’ predictions through error-modulated plasticity in reservoir spiking networks,” the substrate is a \(1{,}000\)-neuron heterogeneous Izhikevich spiking reservoir with about \(80\%\) excitatory and \(20\%\) inhibitory neurons, fixed recurrent weights, sparse Dale-consistent connectivity, double-exponential synapses, and a \(1\) ms integration step. The reservoir itself is not plastic; plasticity resides in the readouts. The “when” pathway uses an error-modulated, attention-gated three-factor Hebbian rule built from presynaptic state \(r(t)\), signed timing error \([y_{\text{when}(t)}-Z_{\text{when}(t)}]\), an attention gate \(G(t)\), and a mask \(P_{\text{mask}}\). The “what” pathway is updated offline at trial end from the post-cue average reservoir state \(\bar r\) [2510.14382]. The model explicitly splits the prediction object into fast and slow learning timescales. “When” learns online within trials; “what” learns offline at trial end. The combined prediction is \[ Z_{\text{prediction object}(t)} = Z_{\text{what}} \times Z_{\text{when}(t)}, \] so identity selects the channel, timing selects the moment, and the product yields a probability-weighted, time-localized expectation. Probability is encoded in readout magnitude rather than in a separate explicit scalar module: higher probability corresponds to stronger readout amplitude near the target window, and lower probability to weaker amplitude [2510.14382]. The operational significance is adaptation under nonstationarity. After a block switch, RMSE briefly rises and then rapidly reconverges; predictions relock to new timing windows and channel amplitudes shift to match new probabilities. The model rapidly adapts to timing changes such as \([30,90]\) ms to \([90,30]\) ms, to probability changes such as \(p(A)\) flipping from \(80\%\) to \(20\%\) or \(100\%\) to \(0\%\), and to combined timing/probability changes across blocks. This is contrasted with a “Gated FORCE” baseline trained by recursive least squares, which can fit stationary regimes but adapts poorly under repeated switches and may need explicit resets or forgetting. The local gated rule, by contrast, updates only synapses implicated by the active channel and time window and preserves drift-free recalibration across blocks [2510.14382]. The model is not uniformly inference-time-plastic. Each trial contains a training phase with cue plus teacher signal and learning enabled, immediately followed by a cue-only testing phase with learning disabled, and the reservoir state is never reset between trials. Thus the strongest inference-time-plastic component is the timing pathway during online operation, whereas identity/probability learning remains a trial-level consolidation step [2510.14382]. A plausible implication is that inference-time plasticity may be computationally most useful when restricted to variables that demand rapid temporal credit assignment. ## 4. Continuous unsupervised adaptation and recurrent self-organization A more fully online formulation is provided by “Time-Integrated Spike-Timing-Dependent-Plasticity,” which proposes TI-STDP as a rule for continuous unsupervised adaptation to sensory input streams. TI-STDP is defined in terms of the last presynaptic spike time \(t_i\), the last postsynaptic spike time \(t_j\), the current time \(t\), and the current synaptic weight \(W^\ell_{ij}(t)\). The post-synaptic event-based scale factor is \[ e^{(t_j - t) / \Delta t}. \] When there is a postsynaptic spike but the presynaptic neuron has been silent, the synapse decays according to \[ \frac{dW^\ell_{ij}(t)}{dt} = -\gamma * e^{(t_j - t) / \Delta t} * W^\ell_{ij}(t). \] When there is presynaptic activity, the update becomes \[ \frac{dW^\ell_{ij}(t)}{dt} = \frac{-\beta}{(t_i - t_j) / \Delta t - 0.5} * e^{(t_j - t) / \Delta t} * (1 - W^\ell_{ij}(t)). \] These are explicitly continuous-time evolution laws rather than event-local increments [2407.10028]. The theorem-level contribution is to establish closed-form synaptic trajectories between spike events. With both pre- and postsynaptic activity, the weight satisfies \[ W(t) = 1 + (W(t_l) - 1)e^{\frac{\beta}{(t_i - t_j) / \Delta t - 0.5}(e^{(t_j - t_l) / \Delta t} - e^{(t_j - t) / \Delta t})}, \] with corollaries stating potentiation for \(0