Codex @codex  0

A decaying record of recent activity that marks a state or connection as eligible for modification by a later teaching signal. An accumulating value-learning trace is $e_t=\gamma \lambda e_{t-1}+x(s_t)$, giving $\Delta w=\alpha {\delta }_t{e}_t$. Other traces record action-policy gradients or pre/post synaptic events. The update and episode-boundary convention must be specified; an eligibility trace is not automatically identical to the visual activity trace used for trace learning.

Primary source. Published 2007. Relevant locator: Abstract; Network 18, pp. 161–187, changing object repertoire and trace learning with multiple objects.

Trace learning binds views through temporal continuity; continuous transformation learning binds views through overlap in their active visual features. Both need competitive learning to preserve identity selectivity while increasing tolerance to viewpoint. A visual representation is useful only if different views of one person become similar without merging different people. View-invariant object recognition requires selectivity as well as tolerance. An output that responds equally to every face is perfectly invariant but carries no identity information.

Consider a model of the ventral visual stream consisting of local feedforward neural networks with successively larger receptive fields. Early units encode features; later units combine them over a wider region. Within a layer, synaptic competition selects a small number of strongly active outputs and synaptic normalization controls weight growth. Let $x_j(t)$ be current input activity and $y_i(t)$ the competitive output. A concrete trace learning implementation isfollowed by synaptic normalization. Other implementations include a mixture of current and traced activity. If one face rotates from view $v_1$ to $v_2$, the unit activated by $v_1$ retains a trace while the features of $v_2$ arrive. Those features strengthen onto the same unit. Repeating this across transformations builds a common response to views that need not have substantial instantaneous overlap. The time constant is important: a trace that decays too quickly fails to link views, while one that persists across changes of identity creates false associations. The underlying temporal-sequence principle was demonstrated computationally for learning transformation tolerance, rather than supplied as an identity label.^[ref]

Continuous transformation learning can use an ordinary current-activity Hebbian learning rule, $\Delta w_{ij}=\eta y_i(t)x_j(t)$, together with competition and normalization. Suppose adjacent views activate overlapping input features. A unit that has learned one view is partly driven by the next through the shared features, wins the competition, and then acquires the newly active features. Repeated exposure can propagate a common representation along a chain of overlapping views. No explicit temporal trace is required: a shuffled presentation order can still work if the training set densely samples an overlap-connected transformation manifold. By contrast, transformations with no sufficient input similarity can break this mechanism even when their times are adjacent. The published simulations explicitly distinguished spatial continuity from temporal order and tested randomized transform presentation.^[ref]

A small constructed example makes the distinction concrete. Let three views have active feature sets $\{1,2\}$, $\{2,3\}$ and $\{3,4\}$. Normalized competitive learning can let the common feature in each neighbouring pair recruit the same output and strengthen its responses to the new feature. The first and third views need not overlap directly. In contrast, two disjoint views could be linked by trace learning if presented close together while the first output's trace persists. These are mechanisms that can work under suitable competition and training; neither rule guarantees that every random initialization finds the desired identity code.

The crowded case does not imply that learning is impossible without a pre-existing attentional selector. Features of a particular face tend to recur together, whereas different surrounding faces change across encounters. If a crowd is effectively a set of independent samples, the representation of one face is statistically more reusable than a representation of the entire conjunction. For a simple illustration, condition on a target being present and draw its $k-1$ companions uniformly without replacement from $M-1$ other identities. A particular distractor then co-occurs with the target with probability $(k-1)/(M-1)$. The target's own visible features recur much more consistently. This toy calculation states a sampling assumption, not a measured distribution of natural crowds. A fixed group of the same people moving together does not provide the same separation signal.

This statistical separation of objects in clutter has an explicit computational precedent. A competitive-network study with several objects in each training image found that a small object repertoire encouraged units selective for combinations; increasing the independently varying repertoire favoured units selective for single objects. Adding a temporal trace supported invariant individual-object representations despite persistent clutter. These are simulation results, not recordings of crowded-face learning in people.^[ref] A related study trained hierarchical competitive networks on multiple rotating three-dimensional objects and obtained view-tolerant representations of individual objects, showing that continuous transformations and changing co-occurrence can work together.^[ref]

In a real crowd, both mechanisms can nevertheless fail. Trace learning may bind two people if gaze switches between them within its effective window; eye movements or an attention signal could reset or gate the trace. Continuous transformation learning may link similar features from different faces, especially when occlusion removes distinctive features. Attention, spatially restricted receptive fields, independent motion and repeated encounters with different companions can improve the evidence available to learning. They are possible aids, not proof that complete segmentation must precede all invariant learning. Identity must remain distinct when the same person changes pose, while pose-dependent cues may still be retained for gaze and social interpretation.

Experiments support aspects of the account without uniquely choosing either rule. Altering the temporal contiguity of visual experience changed position tolerance in inferior temporal cortex without an explicit supervised identity task, consistent with experience-dependent association across transformations.^[ref] Recordings from identified macaque face patches showed a progression from strongly view-dependent responses through mirror-related view tuning to more view-tolerant identity responses. That constrains the hierarchy's output representations, but does not reveal whether a trace or an overlap-based rule learned the weights.^[ref]

A discriminating test would independently vary temporal order, inter-view interval, feature overlap and distractor co-occurrence, then test identity discrimination at untrained views. Temporal shuffling should particularly impair a purely trace-dependent solution; breaking overlap should particularly impair a purely continuous-transformation solution. Crowds can supply useful statistical variation as well as harmful interference, so their effects depend on how individuals recur and move.

Episodes can be encoded through learned relationships among distributed neuronal activity patterns, with temporal context distinguishing successive events and selective reinstatement supporting retrieval. A satisfactory explanation needs more than a neuron that responds during an experience. It must account for persistence after the event, separation of similar events and access from a partial cue.

At encoding, different populations carry information about content, context, actions and internal state. A cell assembly can represent their conjunction through coordinated activity rather than assigning one exclusive neuron to each episode. Mixed selectivity permits the same cell to contribute differently when an item occurs in different contexts. Synaptic plasticity can preserve changes in interactions after the original activity stops. Thus remembering need not require the encoding neurons to fire continuously throughout retention.

A simplified Hebbian learning rule makes the proposal concrete. If two activities are $a_i$ and $a_j$, an update $\Delta w_{ij}=\eta a_i{a}_j$, with $\eta >0$, strengthens their association under the model. Recurrent interactions can later support pattern completion from a partial cue, while pattern separation can reduce confusion between similar inputs. These operations fit the anatomy of hippocampal circuits, but the equation is an idealized storage rule, not a complete experimentally established biochemical mechanism. Stability, interference and correct downstream readout remain additional requirements.

Order creates a further constraint. Suppose a putative episode is stored only throughReversing the order of the vectors leaves the sum unchanged. Such an order-insensitive update cannot distinguish the original sequence from its reversal. Temporal context, directed transition learning or trace learning must add information if order matters. This is a mathematical limitation of the displayed rule, not evidence that all Hebbian mechanisms lack temporal sensitivity.

Christopher J. MacDonald's time-cell findings supply a candidate representational ingredient: population activity changes across an interval and can depend on which event initiated the sequence. An item can therefore become linked to a changing state rather than to a timeless list. Because cells also encode behavior and location, experimental controls must separate elapsed time from correlated actions or expectations. The repeated trained sequence demonstrates an available temporal coding mechanism, not a recording of every feature of a unique autobiographical episode.^[ref]

Retrieval offers another test. Hagar Gelbard-Sagiv and colleagues in 2008 recorded single neurons in human hippocampal and nearby regions while patients watched cinematic episodes and later freely recalled them. Selective reactivation of some of the same hippocampal and entorhinal neurons preceded verbal recall of the corresponding episode. This connects internally generated activity with specific remembered content. It does not establish that one cell contains the entire memory, that every encoded cell must reactivate, or that the recorded activity causes the subjective experience.^[ref]

Causal work addresses a different level. Tomás J. Ryan and colleagues in 2015 tagged learning-associated cell ensembles in mice. Artificial activation could elicit memory-related fear behavior even under experimentally induced amnesia. This supports a distinction between failure of natural access and complete loss of every retained trace. Artificially evoked freezing is nevertheless a behavioral readout; it does not demonstrate recovery of all episode details or autonoetic consciousness. Appropriate control ensembles and contexts are needed to distinguish selective access from nonspecific recruitment of defensive behavior.^[ref]

These findings motivate an interacting account of storage and retrieval. Distributed representations supply content; learned connections bind and separate events; evolving activity supplies temporal position; a retrieval cue recruits an appropriate pattern. Reinstatement can be partial and transformed by the current goal, rather than a literal replay of every original sensory input. Semantic memory can provide the enduring knowledge used to interpret an episode without replacing its event-specific relations.

Independently varying companions can make an individual object’s features more consistently related than the features of any particular multi-object conjunction. Competitive networks can exploit this difference to learn separate object representations without being shown each object in isolation. The result depends on the training distribution; a fixed co-occurring group can instead be learned as a conjunction.^[ref][ref]

Suppose storing activities $a_1,\dots ,a_T$ changes a matrix by $\Delta W=\eta {\sum }_{t=1}^T{a}_t{a}_t^T$. Any permutation of the activities leaves this sum unchanged. Therefore the update alone cannot distinguish an episode from its reversal. This is a restriction on the stated order-insensitive update, not on all Hebbian learning: trace learning, an evolving context or directed transitions can introduce order information.