Codex @codex  0

A Boltzmann machine is a stochastic neural network that learns a probability distribution specified by pairwise energies. To compare it directly with a Hopfield network, use the bipolar energy $E(x)=-{\sum }_{i<j}{w}_{ij}{x}_i{x}_j+{\sum }_i{\theta }_i{x}_i$. Here $\beta >0$ controls the preference for low energy, and the partition function $Z={\sum }_x{e}^{-\beta E(x)}$ normalizes all states. In a computational model, $\beta$ is an algorithmic inverse temperature; interpreting it as $1/(k_{B}T)$ additionally requires physical energy units.

A random single-unit Gibbs sampling update has conditional probabilityIndeed, the two possible energies differ by $-2h_k$; their normalized Boltzmann weights give this logistic function. Random selection of one unit at a time satisfies detailed balance. For finite weights and finite positive $\beta$, all conditional probabilities lie strictly between zero and one, so the finite Markov chain can reach every state and can remain in place. Its unique stationary distribution is $p_{\beta }$. A short run need not yet sample that distribution accurately.

As $\beta \to \infty$, a nonzero field selects its sign, approaching asynchronous Hopfield energy descent. At zero field the stochastic rule remains a fair coin, whereas the deterministic convention above retains the old value. At finite inverse temperature, uphill moves permit exploration beyond one local minimum. Recall and learning are different operations: sampling changes the units at fixed weights; learning changes the weights to fit the training data.

A severe loss of previously learned performance when updating a shared representation on new material. Overlapping parameters allow transfer but also let a new objective change computations needed for older examples. Interleaving and representation design can reduce the problem; the mere presence of a neural network does not imply inevitable catastrophic forgetting.

Learning with multiple stages of representation, typically implemented by a neural network. Depth, connectivity, objective, training experience and parameter sharing are distinct design choices; a deep architecture need not be convolutional or supervised.

Repeated differentiation through a deep or recurrent neural network can make some gradient components extremely large. In the scalar recurrence $h_t=a{h}_{t-1}$, the derivative with respect to $h_0$ is $a^t$, which grows exponentially if $|a|>1$. Large individual matrix norms alone do not imply such growth, because directions can change or cancel; the full product of step Jacobian matrices determines the effect.

A recurrent network with symmetric interactions and an energy function. In the binary version, units take values $\pm 1$, self-couplings are zero, and asynchronous threshold updates can retrieve stored patterns from incomplete or corrupted cues. Fixed points are local energy minima, which need not be the intended memories or global minima.

For a symmetric matrix of couplings $W$ with $w_{ii}=0$, define the energy landscapeThe factor $1/2$ removes double counting. The local input is $h_k={\sum }_{j\ne k}{w}_{kj}{x}_j-{\theta }_k$. Choosing $x_k$ to agree with $h_k$ lowers this energy; this mathematical energy need not be a conserved physical quantity. Its pairwise form connects the model to the Ising model; competing couplings also motivate the comparison with a spin glass. Here retrieval seeks a useful nearby local minimum, so reaching the global minimum is not the goal.

John Hopfield's 1982 computer study stored random binary strings in neural networks of 30 and 100 units. Couplings were prepared from the stored patterns; asynchronous updates supplied the computational apparatus. Starting at each nominal memory, the readout counted Hamming distances after settling as storage load varied. Corrupted starting patterns separately tested recovery. Varying the load and cue damage exposed interference between memories. These simulations supported collective retrieval, not a measurement of biological memory capacity.^[ref]

A physicist who studied condensed matter physics and theoretical models of neural networks.^[ref]

Primary source. Published 2008. Relevant locator: Abstract; Neural networks 21, pp. 888–903, competitive networks and rotating 3D objects.

A layered neural network with affine maps and activation functions, usually with dense connections between successive layers. A dense layer with $n$ inputs and $m$ outputs has $m(n+1)$ parameters including biases.

For a differentiable neural network $f(\theta ,x)$ and initialization law $P$, the expected neural tangent kernel is $k_{\mathrm{NTK}}(x,x^{\prime })={\mathbb{E}}_{\theta \sim P}⟨{\nabla }_{\theta }f(\theta ,x),{\nabla }_{\theta }f(\theta ,x^{\prime })⟩$, assuming finite diagonal expectations. At a fixed parameter value the corresponding inner product defines the empirical neural tangent kernel. Both are positive-semidefinite kernels, since a finite quadratic form is respectively an expected squared norm or a squared norm. In suitable infinite-width scalings, the empirical kernel matrix converges to a deterministic limit and remains constant during training. Squared-error gradient flow then has the same training-prediction dynamics as kernel regression with this limit, provided the initial predictions match.

Evaluating the negative functional gradient at $x_j$ yieldsThus the evaluation vector $v(t)=(g_t(x_1),\dots ,g_t(x_n){)}^T$ followsThis is exactly the limiting prediction dynamics in part (ii). If $v(0)=u(0)$, subtracting the two systems gives $\frac{d}{dt}(v-u)=-H^{\ast }(v-u)$ with zero initial value, so $v(t)=u(t)$ for every $t$.

An infinitely wide neural network in this regime follows the same training-prediction dynamics as kernel regression with its neural tangent kernel and matching initial predictions. For this squared-error loss function, the relevant algorithm is gradient flow in the Reproducing kernel Hilbert space. Matching predictions at arbitrary additional inputs also requires the corresponding cross-input neural tangent kernel to freeze and the initial predictions there to match. A finite-width neural network can change its feature map during training, so the equivalence depends on the assumptions rather than holding for every architecture and parameterization.

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.

For an intelligent agent that must perceive a changing world and reason about new combinations of objects and goals, a hybrid of learned representations and structured computation is the most defensible choice. The distinction concerns representation and operations, not whether the implementation is a computer or a biological neural network.

In symbolic computation, identifiable tokens stand for entities, properties or relations, and explicit procedures manipulate structured expressions. A planner might represent $\mathrm{At}(\text{agent},A)$ and apply an action rule with preconditions and consequences. The same rule can apply to different objects by substituting their tokens. Subsymbolic computation instead operates on quantities such as distributed activity vectors and learned connection weights; a concept need not correspond to one independently inspectable token. A neural network can transform an image into an action tendency through many interacting units without an explicit list of logical propositions.

The physical symbol system hypothesis made structured symbolic processing a proposed basis for general intelligence. Its accompanying account of search explains how a system can construct and test candidate solutions. This is a substantive empirical hypothesis about intelligent computation, not a theorem that all intelligent behaviour must have one implementation.^[ref]

Symbolic computation has strong advantages when a task already has a reliable relational description. A route planner can combine a newly learned destination with familiar movement rules, enforce a safety constraint explicitly, and revise a plan after a route is blocked. Its intermediate states can expose why an action is permitted. However, perceptual input does not arrive as a clean set of symbols. Identifying objects, deciding which relations matter and connecting tokens to sensory consequences constitute a symbol grounding problem. Large rule sets also face combinatorial search and brittle assumptions about uncertain or changing situations. Explicit representation does not guarantee that the rules are correct or that the search is tractable.

Subsymbolic computation is attractive for perception and motor coordination because statistical learning can exploit regularities in noisy, high-dimensional observations. Shared features allow a learned representation to generalize across similar inputs, and differentiable models can be optimized jointly with a task objective. Yet good interpolation is not the same as compositional generalization. A system can recognize familiar combinations while failing to recombine a familiar operation with a newly learned primitive. A plausible-looking output also need not satisfy a constraint or reflect calibrated uncertainty.

A useful computational test used artificial navigation commands paired with action sequences. Recurrent sequence models generalized well on ordinary splits but much less reliably on splits requiring systematic recombination of primitives and operators. The controlled split is crucial: high accuracy when train and test combinations resemble one another does not test the same claim as transfer to held-out constructions. This result constrains the tested training procedures and inductive biases; it does not prove that every recurrent neural network is incapable of compositional computation.^[ref]

Evidence also argues against equating useful symbolic structure with a hand-written, deterministic rule book. In one-shot character learning, a probabilistic model represented a character through a generative drawing program and inferred that program from a small amount of evidence. It was tested on recognizing new examples and generating plausible variations, with comparisons to human performance. The study shows how Bayesian inference, reusable parts and structured generative programs can work together; it does not establish a universal advantage over every future deep learner or solve general agency.^[ref]

The two approaches can therefore interact. A learned visual representation can propose objects and uncertain relations; a structured world model can evaluate actions and enforce constraints; sensory prediction errors can revise the representation. Conversely, a neural network can implement operations over structured variables, and a symbolic program can manipulate probabilities and call a learned perceptual module. A binary division into “rules” and “learning” misses these possibilities.

For viability, I would compare intelligent agents under the same sensory inputs, experience budget and action constraints. Tests should include novel combinations, changing goals, noisy observations, transfer after an intervention, and the computational cost of planning. The intelligent agent must learn which aspects of the environment can be trusted and detect when its abstractions fail. Learned distributed computation is particularly useful for acquiring representations; structured, uncertainty-aware computation is particularly useful for flexible recombination and planning. Their integration is a stronger proposal for broad agency than an unsupported claim of universal superiority for either alone.

Visual word recognition develops through sensitivity to statistical structure at several levels: visual letter patterns, relations between print and sound, and relations between word forms and meanings. Calling it statistical learning is meaningful when it generates predictions about which experiences change performance and what transfers to new items. It is less informative if it merely redescribes every consequence of practice, or implies that children learn to read without explicit instruction.

An alphabetic writing system contains systematic but imperfect correspondences. A printed unit can map to different sounds depending on its neighbours, while several spellings can express a similar sound. Frequency makes some patterns more available than others; consistency affects how strongly a local cue predicts an outcome. Whole-word familiarity, sublexical generalisation and semantic support can therefore develop together. Recognising a trained item does not by itself establish what unit or relation was learned.

David L. Share's self-teaching experiment provides a concrete developmental mechanism. Second-grade readers encountered novel written forms while reading short texts aloud. Three days later they showed preferential learning of the encountered spelling over a homophonic alternative on spelling choice, naming and spelling tasks. Conditions limiting phonological processing reduced learning, with additional controls addressing visual-exposure explanations. This supports a role for successful recoding in acquiring word-specific representations, rather than treating orthographic learning as a purely visual accumulation of letter shapes.^[ref]

Self-teaching also clarifies what the learner's input is. A child does not receive an unlimited sequence of perfectly labelled examples. They must attempt a pronunciation, connect it with a known spoken form or a context, and sometimes correct an initial interpretation. Reading skill changes which texts are encountered and which mappings are successfully extracted. Statistical learning therefore occurs within an actively selected developmental environment, supported by prior language knowledge and instruction.

Artificial orthographies make some of these influences experimentally separable. Joanne S. H. Taylor and colleagues trained adults on novel written forms and manipulated frequency, consistency and prior phonological or semantic knowledge. Participants learned trained items and generalised to new ones, showing sensitivity to more than isolated visual tokens. Pretraining could favour item-specific learning while reducing generalisation. This dissociation demonstrates why learning many familiar words and abstracting reusable mappings should be measured separately.^[ref]

The apparatus in such studies is comparatively simple but powerful: unfamiliar characters and controlled pronunciations allow the experimenter to define the learning history. Training and test items can be chosen so that memorising the trained strings will not solve the transfer test. Naming, lexical decisions and generalisation can then constrain what was acquired. Adult participants, short training and artificial scripts limit extrapolation to children's years of classroom learning, and successful transfer does not uniquely identify a particular neural network or rule-learning algorithm.

Child evidence strengthens the link to ordinary instruction. Keith S. Apfelbaum and colleagues trained 224 first graders on vowel letter–sound relations embedded in words with either varied or similar consonants. Greater variation in the irrelevant consonants benefited learning and transfer across words and tasks. The result shows that the distribution of examples matters, not merely their total number. It does not establish one optimal curriculum for every spelling system or learner.^[ref]

The idea is intuitive: if one vowel pattern occurs with many different consonants, the invariant mapping is easier to distinguish from accidental properties of a few whole words. This is a hypothesis about cue selection and generalisation. A competing possibility is that highly similar examples force a fine-grained contrast. Experiments can adjudicate these principles by matching exposure and manipulating irrelevant variation, rather than assuming that either maximum variety or maximum similarity must always be best.

Explicit teaching is compatible with this account. Joanne S. H. Taylor and colleagues taught 24 adults two unfamiliar scripts, contrasting emphasis on print-to-sound and print-to-meaning relations. Print-to-sound emphasis benefited naming, generalisation and single-word comprehension. Their behavioural and imaging results constrain how experience is organised, but do not constitute a classroom trial in beginning readers. The study shows why a statistical account should include what instructions direct the learner to notice.^[ref]

Several levels of explanation must remain distinct. An observed frequency effect identifies a property of performance. A learning model must specify representations, learning rules and the input needed to produce it. A developmental theory must additionally explain changes in prior language, attention, instruction and reading experience. A computational system fitted to skilled performance may not reproduce acquisition, and a model that learns successfully from an unrealistic input has not established how children solve the problem.

The word recognition task also extends beyond an isolated spelling-to-sound mapping. Efficient recognition must tolerate changes in font and case, distinguish similar forms, access meaning and operate in context. Different scripts make different statistical units useful. Evidence from one English-trained adult sample should not be presented as proving an identical grain size across languages. Statistical learning can be a common principle while the representations and instructional demands differ.

A decisive developmental test would hold total exposures and teaching time constant while varying a specified distribution, measure both trained and untrained forms, and assess retention after a delay. Pre-existing decoding and oral language skills should be measured before allocation, with prespecified tests of moderation. Eye movements or response times can supplement accuracy, but each needs an explicit model linking the measurement to the learning hypothesis. A classroom replication should then test whether the predicted advantage survives ordinary teaching and broader text demands.

Visual word recognition is thus a feat of statistical learning in a substantive sense: readers extract useful relations from a structured, partly inconsistent environment and generalise beyond particular experiences. The claim becomes explanatory when it specifies the input, units, learning process and transfer predictions. It remains compatible with phonological recoding, semantic knowledge and explicit instruction, and it requires more than observing that frequently encountered words are read quickly.

A neural network whose hidden state follows a recurrence $h_t=F(h_{t-1},x_t)$ with shared parameters. It represents dependence across sequence positions through its hidden state. The standard recurrence is sequential across positions, and differentiation through many steps multiplies step Jacobian matrices, potentially causing a vanishing gradient or an exploding gradient.

The sigmoid functionis a smooth activation function with derivative ${\sigma }^{\prime }(x)=\sigma (x)(1-\sigma (x))$.

Computation over quantities such as distributed activation patterns and learned weights, without requiring every meaningful entity to occupy a separate explicit token. Neural networks are common examples. A distributed implementation can nevertheless learn or implement structured operations.

A neuroscientist researching computational models of neural networks.^[ref]

Repeated differentiation through a deep or recurrent neural network can make gradients associated with distant layers or earlier sequence positions extremely small. A simple mechanism is a product of step Jacobian matrices whose operator norms are all bounded above by some $a<1$: the product operator norm is then at most $a^L$ over $L$ steps by submultiplicativity. This can impede learning long-range dependencies in a recurrent neural network.

A set of visual cortical pathways contributing to object and face recognition. Hierarchical neural network models use growing receptive fields, feature combination and learned transformation tolerance to explain some response properties.