Neural Representation Geometry

We develop a geometry-based capacity theory for exact-key retrieval in compressed finite-feature Hebbian associative memory. For random or approximately isotropic values, retrieval interference separates into finite-feature noise, which decreases with feature dimension, and structural interference, which is determined by squared kernel overlap among stored keys and persists in the infinite-feature limit. This yields a fit-free prediction of retrieval quality, reveals a geometry-dependent capacity ceiling, and predicts the feature budget required for a target retrieval quality. When stored values are correlated, we show that retrieval depends jointly on the key kernel and value Gram matrix, and derive finite-feature approximations that account for this interaction. We validate the theory on synthetic, visual, and medical-image representations. Overall, the framework links representation geometry directly to memory capacity and distinguishes when performance can be improved by increasing the feature budget and when the representation itself must be changed. Across these settings, the predicted retrieval curves closely match empirical behavior and correctly identify changes in the preferred memory design.

Geometric alignment scores based on Gram determinants provide a compact way to model higher-order consistency among modalities, yet how such scores respond to modality degradation is poorly understood. This paper asks whether the response of a multimodal geometric score is determined primarily by the magnitude of the perturbation-induced displacement. Using frozen cohorts from MSR-VTT (N=878) and DiDeMo (N=980), we apply controlled video blur and audio noise and analyze the response in the relational geometry on which the score is defined. Displacement magnitude explains at most 15% of the out-of-sample variance in the absolute response, and magnitude-matched pairs respond systematically differently, so scalar magnitude does not organize the response. The closed-form first-order expansion of the Gramian volume yields the Directional Geometric Response (DGR): the projection of the displacement onto the local volume gradient, which jointly captures the clean operating point, displacement magnitude, and displacement direction. The absolute first-order DGR term explains the observed response with out-of-sample R^2 of 0.838-0.969, matched-magnitude ranking accuracies of 0.864-0.963, and response-sign accuracies of 0.909-0.989, whereas the tested direction-free alternatives remain weak or unstable under the corresponding evaluation protocols. A pre-specified gain-normalization candidate, V/(g_V+eps), fails its predictability and clean-order gates. DGR uses the observed degraded-state displacement and is therefore an explanatory quantity, not a deployment-time predictor: geometric response depends on where the representation operates, how far degradation moves the relational geometry, and in which direction it moves.

Learned feature representations in neural networks often contain structure beyond that directly used by the final task output. We study this structure using $\textit{passive probes}$ that apply fixed, untrained, property-independent projections to representations as they evolve during training. We motivate this approach through the task of prediction on $S^2$ where equivalent vector and Hermitian parameterizations reveal an additional loss-invariant trace coordinate. This motivates a general construction in which fixed random projections serve as observers of learned features. Because the observer is loss-invariant and independent of the property being studied, changes in accessibility reflect changes in the representation relative to the fixed observer rather than adaptation of the observer itself. We show that ensembles of passive probes can directly reflect task-relevant information such as target alignment. Under our constructions, the accessibility of eventual difficulty evolves differently across tasks. It increases during training in the regression tasks of surface-normal estimation and image inpainting but remains near its initial level in image classification. Comparisons with learned linear probes further show that recoverability and passive accessibility can evolve differently during training. Together, these results show how passive probes can separately characterize changes in representation geometry and the accessibility of eventual task difficulty.

Explaining generalization and training dynamics in neural networks remains a challenge, and various approaches have been developed to study different aspects of these phenomena. In this paper, we introduce the idea of {\em magnitude potential} -- a quantity based on the theory of metric magnitude -- that reflects how well an arbitrary point is represented by a given set. We find that this basic quantity can be applied to examine various features in neural generalization. The ratio between the magnitude potential with respect to a class and with respect to the entire data, computed at the logit layer, is informative of the representation of the point. In experiments, these ratios for individual training points are found to be correlated with the Feldman memorization scores. Magnitude potential ratios aggregated across points detect structural changes in the decision boundaries and provide a geometric indicator of grokking in modular arithmetic. Although the magnitude potential ratio and neural collapse are both closely associated with intra-class and inter-class geometric structure, the magnitude potential ratio remains informative even when neural collapse is explicitly suppressed.

Vision-language models (VLMs) have emerged as powerful candidates for universal vision backbones, with representative architectures including autoregressive (AR) models and diffusion transformers (DiTs). Yet, adapting them efficiently for all-in-one low-level image restoration remains a challenge. Crucially, the field lacks an understanding of how VLMs organize hidden-layer representations and whether these structurally distinct paradigms share a common geometric organization for pixel-level perception. Such shared organization is a prerequisite for building highly transferable, unified restoration VLMs and adapters. In this paper, we systematically investigate representational similarity across 24 low-level tasks spanning 5 categories. We propose GeoSim, a unified four-level framework that analyzes task-conditioned representations from global similarity, local geometry, sparse feature decomposition, and topological verification perspectives. Our formulation applies to the analysis of hidden states in AR models and feature maps in DiTs across same- and cross-task/model settings. Our results reveal the organizing principles of low-level visual representations while exposing their limits in cross-task and cross-model agreement. Ultimately, GeoSim provides an interpretability lens for probing latent transferability in low-level vision and diagnosing model limitations in task- or model-specific scenarios.

Understanding how learned representations respond to finite input changes is important for characterizing their sensitivity, invariances, and robustness. Yet existing geometric analyses are predominantly local and describe only infinitesimal perturbations. We introduce a scale-resolved statistic that compares an encoder's measured feature displacement with its local linear prediction as the perturbation magnitude increases. Across diverse image encoders, we discover a characteristic plateau-rise-peak-decay profile, which we call the bump. The bump is absent at initialization, emerges early during standard training, and does not form under randomized labels or random-noise inputs. Its shape also varies with the training distribution and robustness objective. These results establish departures from local geometry as a signature of how encoder representations are shaped by learning.

Representations are routinely used across machine learning, psychology, and neuroscience to draw inferences about the computations of biological and artificial systems. Such inferences presume a meaningful link between representational geometry and the computation being performed. For artificial neural networks, however, the extent to which function constrains representation remains unclear. One key obstacle is that these networks admit parameter symmetries: changes in parameterization that preserve function exactly while reshaping representational geometry. Here, we show that a broad class of parameter symmetries acts on representations through just three primitive feature transformations: addition, duplication, and scaling. This feature-level characterization yields a closed-form decomposition of representational geometry into essential and auxiliary components, which makes precise how degeneracy in representational geometry can grow with overparameterization even when function is held fixed. Finally, we show that implementation-level selection rules can resolve this degeneracy, yielding identifiable geometries in which features are weighted according to their contributions to the network's function. Together, our results delineate when representations can support inferences about computation, and when they cannot.

A foundational principle of connectionism is that perception, action, and cognition emerge from parallel computations among simple, interconnected units that generate and rely on neural representations. Accordingly, researchers employ multivariate pattern analysis to decode and compare the neural codes of artificial and biological networks, aiming to uncover their functions. However, there is limited analytical understanding of how a network's representation and function relate, despite this being essential to any quantitative notion of underlying function or functional similarity. We address this question using analysable two-layer linear networks and numerical simulations in non-linear networks. We find that function and representation are dissociated, allowing representational similarity without functional similarity and vice versa. Further, we show that neither robustness to input noise nor the level of generalization error constrain representations to the task. In contrast, networks robust to parameter noise have limited representational flexibility and must employ task-specific representations. Our findings suggest that representational alignment reflects computational advantages beyond functional alignment alone, with significant implications for interpreting and comparing the representations of connectionist systems.

Semantic-preserving transformations can induce substantial motion in learned representations, while small changes may strongly affect model predictions, raising a basic question: what local metric best captures semantically consequential variation? We propose Fisher-induced invariant representation geometry (Fisher-IRG), which measures local representation directions through their predictive sensitivity. Around each representation, we construct semantic-preserving and semantic-changing neighborhoods, aggregate their local Fisher information, and recover invariant directions through a contrastive generalized eigenvalue problem. Controlled displacement analyses first show that comparable Euclidean motion can have substantially different predictive consequences, supporting the need for a predictive geometry. Across language and vision models, Fisher-IRG yields stronger semantic-versus-nuisance predictive selectivity and generally more reproducible subspaces than covariance-based geometry, while recovering systematically distinct local directions. Representation interventions further localize semantic effects to the Fisher-derived subspace, and held-out separation and retrieval show that the recovered geometry generalizes beyond the discovery neighborhoods. These results support Fisher-IRG as a principled framework for characterizing local invariant representation geometry.

Representations continually change as a network learns new tasks. We ask whether early representational changes naturally form a geometric structure that continues to shape later learning. We identify a low-dimensional subspace of early representation drift, which we call a scaffold, and test whether it is reused across subsequent tasks. Across four pretrained visual encoders and two datasets, later representational changes consistently favor this early-defined subspace over matched random alternatives. This reuse is history-dependent: when networks experience different early tasks but identical later training inputs, each network preferentially reuses the scaffold induced by its own learning history. The same preference appears in individual optimizer updates, even though the network's dominant local response directions shift away from the original scaffold. Finally, constraining motion within the scaffold slows new-task acquisition more than matched random constraints, while effects on old-task retention are less consistent. In summary, these results suggest that early experience leaves a persistent geometric imprint on how neural networks adapt to future tasks.Code is available at https://github.com/YuantaoDeng/latent-scaffold.

The projection of queries and keys are central to the attention mechanism in Transformer architectures. While they are mathematically symmetric, they play different roles in attention mechanisms. The question of whether there is an effect from their functional distinction on their geometric development in training remains unanswered. We investigate the problem through the training of small GPT-like Transformers on character-level WikiText-103 for three different depths (4, 6, and 8 layers), three types of initialization for queries and keys, and four random seeds, resulting in 36 runs and 54 trajectories of average layers across seeds. We track the effective dimensionality of those layers using participation ratios and discover that effective dimension of queries expand while keys shrink, and that $PR_Q - PR_K$ is positive in all trajectories studied. In connection to attention, the shrinking of keys leads to a narrower spectrum of $QK^\top$ and more peaked attention weights. In order to determine if this connection is causal or coincidental, we directly control the spectrum of keys during training across five seeds: restricting it to make it shrink sharpens the attention with high directional confidence, while keeping it constant to the level of initial dispersion makes attention softer. Additional token-level checkpoint analyses show that the monotonic paired-contrast trend is not universal across pretrained families, but survives as an early-training regime that later decays over a full pretraining run, and the link between interaction-rank geometry and attention entropy remains visible in several models.

We study the knowledge matrix of a trained feedforward network as a higher representation of its inputs. A network is a pair $(W,f)$, a thin representation $W$ of its quiver and an activation $f$; its function factorizes through the space of quiver representations, each input $x$ inducing a representation, and the knowledge matrix $M(x)\in\mathbb{R}^{C\times(d+1)}$ is the contraction of that representation to one matrix whose rows sum exactly to the logits. At one trained network we ask what determines it, what it is invariant to, what it determines, and what its geometry measures. Under (LCS), a locally constant slope diagonal, as for ReLU, the matrix at a regular input is a function of the realized germ; its stabilizer among encodings regular there is exactly the germ stabilizer at inputs with no vanishing coordinate, neuron permutation a special case; and it recovers the germ, whereas hidden activations, gauge-covariant and germ-incomplete, are not enough. Under (LCS) it equals per-class gradient$\times$input plus an exact aggregate bias attribution, grounding it in attribution theory and computing it by $C$ vector-Jacobian products instead of probing. The fixed shape gives an alignment-free per-sample distance between ResNet-152, DenseNet-121 and GoogLeNet; the row-sum identity gives an exact visible/invisible displacement decomposition whose unit-free coherence $A=(d_Ψ/d_M)^2$ puts adversarial germ motion at median $A\le 0.23$, with an attack-family ordering concordant across six architectures (Kendall $W=0.921$; $0.97$ on the three networks at full scale). Two honest negatives: on AlexNet/CIFAR-10 penultimate features win 5 of 6 detectors and all 16 attacks, and a matrix-direction counterfactual fails 0/54.

We introduce NeuralSRNF, a novel framework for the statistical shape analysis and generation of genus-zero 3D and 4D objects that undergo nonrigid deformations. Traditional methods rely on complex and computationally expensive nonlinear elastic metrics that measure bending and stretching. Recent advances in elastic shape analysis achieve computational efficiency by mapping input 3D shapes to the space of Square Root Normal Fields (SRNFs) where the L2 metric approximates the partial elastic metric, significantly facilitating the process of computing geodesics and summary statistics. SRNFs, however, are not invertible, and the numerical algorithms used to map SRNFs back to the original space of surfaces remain computationally very expensive and often lead to approximate results. This paper addresses this fundamental SRNF inversion problem using a novel neural representation, termed NeuralSRNF. Unlike the commonly used numerical SRNF, NeuralSRNF is (1) continuous, and thus resolution-agnostic, enabling full functional shape analysis, (2) more accurate, and (3) computationally more efficient as it can compute inverse SRNF maps along a geodesic path in less than 3 s compared to over 10 min for the numerical SRNF. We demonstrate, using various datasets, the utility and efficiency of the proposed NeuralSRNF in multiple elastic 3D and 4D shape analysis tasks such as geodesic computation, deformation transfer, statistical summaries computation, and 3D shape generation. We show that it outperforms competing methods on most evaluated datasets and metrics by a wide margin in both accuracy and computational efficiency. The source code and additional results are available at https://awaisnizamani16.github.io/awais/NeuralSRNF/.

The Platonic Representation Hypothesis suggests that increasingly capable models converge toward shared representations. Recent work narrows this claim to shared local neighborhood relationships, finding that capacity-dependent trends in several global similarity measures largely disappear after calibration. We challenge this interpretation by showing that prior local-global comparisons confound structural scale (local versus global) with what is compared: relational structure, defined by which samples are related, versus metric geometry, characterized by quantitative relations such as distances, similarities, or correlations. To disentangle these factors, we construct a controlled $2\times2$ framework that evaluates both relational structure and metric geometry at local and global scales. We introduce $H_0$ skeleton overlap as a global counterpart to mutual $k$-nearest neighbors, together with matched distance-aware variants. Across vision-language models, relational structure exhibits robust representational convergence at both scales after calibration, whereas increasingly stringent distance agreement substantially weakens alignment and progressively flattens the capacity-dependent trend. We further extend the analysis beyond ambient Euclidean geometry by evaluating distance agreement under a Riemannian metric approximation and recover the same structure-geometry pattern. The pattern is also reproduced in video-text representations. Together, these results show that relational convergence extends beyond local neighborhoods to global spanning structure, whereas metric geometry exhibits substantially weaker convergence.

Angular-margin speaker encoders are widely used in voice conversion, yet the geometry of their classifier prototypes remains poorly understood. We analyze ECAPA-TDNN classifier prototypes as points on the unit hypersphere and characterize their organization using rotation-invariant angular statistics together with global and local effective dimensionality measures. Our analysis shows that standard training can induce angular concentration and a substantial reduction in effective dimensionality. To address this, we investigate two geometric regularization strategies (hinged Riesz log-energy and effective-dimension maximization) applied to classifier prototypes to encourage more uniform hyperspherical coverage. The resulting prototype sets exhibit higher effective dimensionality and improved isotropy, with configuration-dependent effects on speaker-recognition performance. When the corresponding ECAPA-TDNN models are used as speaker encoders for Fast-VGAN, the regularized systems also exhibit improved robustness in zero-shot voice conversion, particularly for previously unseen speakers.

How neural representations preserve the structure of input changes connects representation analysis with internal intervention. We study operable representational content through compatible actions of reference transformations on neural features. We characterise when a transformation descends through an encoder, and give a linear setting in which the defect is governed by the transformation's demand for discarded information, measured in the metric the representation induces. On a rectifier the failure to realise a transformation has two distinguishable sources --- what the source region has already made unrecoverable, and what it costs to satisfy every region the transformation visits with one operator --- and for a \textit{measured} harmonic carrier the same question has a closed answer: a linear realisation exists exactly when the retained harmonic blocks are invariant under the action. Using colour as the in-depth instance, we find that hue orbits in frozen visual features concentrate 84--88% of their energy in the first two harmonics with rotation planes shared across shapes, that this organisation is substantially inherited from input and architecture and is reshaped by training and depth, and that the measured structure supports prediction, transport from new starting states, and composition --- with global and local realisations differing sharply in which they achieve. Guided by the measurements, we construct a compact interface whose rotation action is fixed by the structure and never fitted: it reads hue zero-shot at 3.4$^\circ$ median error on unseen shapes. Theory, structural measurement, and construction together establish transformation laws as a concrete object connecting the understanding of neural representations to their design.

Long-term memory is becoming a core substrate for personalized AI, yet most systems still represent personalization as discrete records in a largely static latent space, accessed under one global similarity notion. For data mining, this creates a mismatch: the evidence is a temporal event stream, while the dominant abstraction is a searchable record set. We argue that long-horizon personalization should instead model memory as a user-specific dynamical state space with locally heterogeneous geometry. Geometry here is a computational language, not a literal claim about cognition: it captures stable versus volatile regions, variable-rate drift, heterogeneous neighborhoods, and uncertainty about current user state. Profiles and isolated events remain useful as points, but interaction, feedback, and elapsed time induce trajectories. Memory access then becomes trajectory-conditioned reconstruction of the relevant user state, not only nearest-neighbor lookup.

Full-context neural visual geometry is impractical for thousands of images, while sequence-based chunking poorly captures irregular non-local overlap in multi-sequence aerial collections. We present Graph-Guided Neural Visual Geometry for Aerial Registration (G3AR), a graph-guided framework for scalable dense neural geometry. Before local inference, G3AR builds a geometrically verified image-proximity graph that guides bounded overlapping chunks and induces a chunk graph whose maximum spanning tree defines alignment topology. Compatible backbones process chunks independently; shared-image predictions then estimate three-dimensional similarity (Sim(3)) transforms that register local cameras and geometry in a common frame. Across four real aerial scenes, G3AR improves pose error and runtime in matched VGGT- and Pi3-backed comparisons, while its DA3 variant achieves the lowest pose error among evaluated neural-geometry methods.

Transformer representations evolve through learned additive transformations that either preserve their current direction or redirect it. We study this evolution as a functional geometry, decomposing learned updates into parallel and perpendicular components. Across pretrained models, we find substantial parallel components beyond the residual identity path. We then apply the decomposition in two spaces: to attention and MLP updates relative to the hidden state, and to attention value aggregation relative to the current token's value. Targeted edits reveal a strongly space-dependent asymmetry: exclude-self value-space parallel manipulation is markedly more robust than residual-space and perpendicular counterparts, preserving the direct self message while scaling only the non-self aggregate. The same decomposition gives a component-resolved description of compression-induced update error: perpendicular error separates compression methods more clearly than parallel error. Extensive experiments further demonstrate that full-aggregate parallel suppression during from-scratch pretraining lowers validation-loss trajectories and improves downstream averages, with the value-space variant strongest. Together, these results connect representation geometry to editing robustness, compression diagnosis, and training-time intervention. Code is available in the project repository.

Neural networks can leverage feature superposition to encode more concepts than dimensions, but cross-feature interference constrains the linear accessibility of simultaneously active features. By framing linear accessibility as a compressed sensing problem, we derive high-probability bounds for fixed supports under subgaussian noise, proving the sufficient dimension scales linearly ($d=O_{\varepsilon}(k \log m)$) rather than prior worst-case quadratic limits. We then validate these bounds across system parameters through Gaussian-tail approximations. These results quantify the geometric constraints of the linear representation hypothesis, providing a framework for evaluating sparse autoencoders, compositional generalization, and neural interpretability.

Class disentanglement (the separation of a representation's class-conditional point clouds along depth and over training) is usually read off descriptive curves. We measure it as certified topological interaction between labeled point clouds, using the recently introduced Intersection Euler Characteristic Profile: the Euler characteristic of the overlap of the clouds' ball unions as a function of scale, computed by one Alpha-complex sweep with no boundary-matrix reduction. Every number carries a test: exact permutation tests in both directions, a guarded separation certificate, and a paired test for the comparative claims applications make. Across 111 trained networks and 52,650 certified measurements, disentanglement is depth-graded and concentrated in the first epochs, and interaction quotients rank class pairs by confusability (Spearman rho=0.83), on par with cheap separability statistics. In a 96-model factorial population, augmentation is the one training choice that separates classes relative to chance; weight decay compresses the overlap without separating, and depth and width do nothing. The structural finding is one only a k-fold statistic can pose: the joint entanglement of a class triple sits below that of its strongest pair in 97% of triple-layer cells and 99.5% of deep cells, far below a measured null floor, in vision encoders and frozen language models alike. This pairwise dominance is a regularity, not a law: expected from the nesting of overlaps but not forced by geometry, present at initialization and in raw pixels, and manufactured in the last stage alone when a network memorizes random labels. The unnormalized profile mass predicts test accuracy (R^2=0.94), the quotient does not, and neither beats a linear probe. One lesson is reported in full: the paired test must use a scale-free statistic, or it certifies feature-norm dynamics as disentanglement.

Existing theories derive neural scaling from data geometry or a specified data-model spectrum, but systems trained on the same data can scale differently when architecture or optimization changes the representations they can efficiently reach. We introduce Coupled Scaling, a task-conditioned framework in which finite-budget scaling depends on the relation between task structure and the geometry accessible to an architecture-optimization system. In a solvable mode-truncation model, loss separates into target energy outside architectural support and an unresolved supported tail. For an arbitrary priority order, the residual lies between the best-N supported tail and the tail beyond the largest completed high-value prefix. If the cumulative-tail and coverage log-rates are $γ_{A,T}$ and $ρ_{A,O,T}$, the residual exponent lies in $[ρ_{A,O,T}γ_{A,T},γ_{A,T}]$. Under bounded off-prefix gain, the completed prefix is rate-determining and $α_{A,O,T}=ρ_{A,O,T}γ_{A,T}$; for $a_{A,T,j}\asymp j^{-b_{A,T}}$, this gives $α_{A,O,T}=ρ_{A,O,T}(b_{A,T}-1)$. A fixed-kernel specialization derives the training-time exponent from the near-zero tail of a task-weighted spectral measure defined independently of the loss fit. The framework separates architectural support from finite-budget acquisition and motivates two tests: static task-relevant geometry should track loss at a common budget, while multiscale geometry should track coupling-specific exponent ordering, including reversal across contrasting tasks. An audit of released emergence trajectories identifies the controls needed for a direct factorial test that measures geometry separately from the scaling fit.

Generative models for implied volatility surfaces must produce outputs that satisfy static no-arbitrage constraints. We study these constraints in latent space. For a fixed generator, we assign each latent code a scalar margin determined by the no-arbitrage conditions of the generated surface. The codes with nonnegative margin form the admissible latent set. We establish conditions under which strictly admissible codes remain admissible under small perturbations and the boundary of the admissible set is characterized by zero margin. For regular boundary components, we formulate a level-set equation whose local dynamics are directed toward the zero-margin set. The analysis treats the generator as a map from latent variables to surfaces and is therefore not restricted to a particular architecture. It applies to variational autoencoders, generative adversarial networks, and other generative models with a deterministic realization map. Numerical tests recover known boundaries in analytic examples. Experiments with a variational autoencoder trained on Heston surfaces show that similar reconstruction errors can correspond to different admissible regions and that the latent prior may be concentrated inside such a region. The computed boundary can also be used to modify latent codes that generate violating surfaces.

Pretrained vision embeddings are widely used to model how people appraise urban scenes, and they are validated almost entirely by how well they predict human ratings. Accurate prediction shows that an embedding contains the information needed to recover the ratings. It does not show that the embedding arranges scenes as the human visual system does, which is assumed when distances or dimensions in the embedding are read as perceptual. Using openly released EEG recorded while 63 adults viewed and rated street scenes of Berlin, we compared the neural representational geometry of the scenes with the geometry of pretrained vision models that differ in training objective and size, of simple image descriptors and of the ratings themselves, each relative to a noise ceiling given by the agreement between participants. No feature space reached more than about half of the noise ceiling, which corresponds to about a fifth of the reliable variance in the neural geometry. A descriptor of oriented edge energy reached the level of most pretrained models while capturing a different part of the neural geometry, and deeper layers corresponded to later neural responses. The same embeddings predicted held-out ratings well, up to r = 0.87, yet prediction accuracy and neural correspondence were not reliably related across models, and the best predictor was among the least aligned. Predicting how a street is appraised is therefore weak evidence that a model represents the street as the brain does.

A token-embedding table holds a hub of short rows near its origin, and we show that this cluster biases what nearest-neighbor intrinsic-dimension (ID) estimators report. Because of the concentration of measure, a token is closer to the central cluster than to any other token, so its first two neighbors are both hub rows at nearly the same distance. As a result, the ID estimators such as TwoNN return a dimension far above the real ID. Measured one token at a time, dimension is a heavy-tailed distribution. Measured on the full vocabulary, it grows with the model's parameter count. However, when we remove the hub, the heavy tail disappears and the measured dimension collapses to a narrow range for eleven models, from GPT-2 to models such as K3 and GLM-4.7. The hub acts as a switch: a few hundred rows are enough to fully inflate the estimate. We reproduced an experiment stating that the intrinsic dimension (ID) of Pythia's token-embedding table grows with the parameter count, from $27$ to $122$ between 160M and 12B parameters. We show that this result disappears when the hub is removed: the table then reads $10$ to $17$ at every size. The hub contains a subset of the population that under-trained-token detectors flag, but on Pythia the hub that we detected and removed as a whole was updated during training: what seem to characterize these rows is simply their length, not an absence of updates. Finally, we show that normalizing the rows instead of removing them gives the same lower reading.

Joint-embedding predictive world models support planning through latent predictions, but unconstrained joint training can collapse distinct observations to identical embeddings. Two prominent strategies for avoiding collapse are to inherit pretrained features, as in DINO-WM, or to learn representations end-to-end with anti-collapse regularization, as in LeWorldModel (LeWM). Yet avoiding collapse does not ensure that latent distances distinguish outcomes in ways that matter for the task. In object manipulation, for example, success depends on the object's position and orientation relative to the goal. Such task-relevant state information can remain accurately decodable while barely influencing latent distance. The resulting planning cost may fail to reflect how close a predicted outcome is to the task goal. In this paper, we propose SCALE (State-CAlibrated Latent Embeddings), a method that correlates sampled pairwise latent distances with distances in task-relevant state space. Added to LeWM's existing objective, SCALE preserves its architecture, requires privileged state only during training, and adds no planning-time computation. We show that SCALE improves planning success over LeWM across manipulation and navigation tasks with multiple solvers and provide a comprehensive analysis of how SCALE reshapes representation geometry to support planning.

Deep representation learning has primarily focused on how features evolve across network layers, while largely overlooking the structured geometry embedded in network parameters. We introduce a dual-manifold perspective in which each convolutional layer contains two coupled geometric spaces: a Kernel Manifold induced by convolutional filters and a Data Manifold characterized by intermediate feature representations. Because these manifolds share the same channel space, parameter geometry can provide complementary structural information to guide feature evolution. Based on this insight, we propose Kernel-Guided Feature Transform (KGFT), a lightweight module that derives a geometric guidance matrix from the kernel Gram matrix and uses it to transform the covariance structure of feature representations. Unlike conventional attention mechanisms that reweight feature responses, KGFT explicitly reshapes feature relationships by transferring geometric information from the kernel manifold to the data manifold. To accommodate network hierarchy, we further introduce Exploit and Explore modes with a depth-aware scheduling strategy and a learnable guidance strength that adaptively controls the contribution of geometric transformation. This design promotes geometric alignment in shallow layers while encouraging feature diversity in deeper layers, without imposing excessive constraints on representation learning. Theoretical analysis establishes the validity of the proposed transformation and characterizes its effect on feature covariance. Extensive experiments across CNN- and Transformer-based architectures, including ResNet, ViT, and LLaMA-7B, demonstrate consistent improvements on image classification and arithmetic reasoning tasks, validating the generality and effectiveness of kernel-guided dual-manifold representation learning. Code will be publicly available.

How many directions does a neural representation use to encode a concept? A common answer repeatedly erases probe directions and reports the stopping count or cumulative removed rank. We show that both quantities can change under an information-preserving invertible reparameterization, so neither is intrinsically a concept dimension. We distinguish model-defined population quantities (generating dimension, sufficient linear dimension, and minimum guarding rank) from procedure-defined quantities such as stopping count and cumulative edit rank. In a population Gaussian construction, an invertible shear preserves the prediction problem and all three quantities, yet changes the cumulative Euclidean erasure count from one to two. The separation holds for Moore--Penrose ordinary least squares and every finite nonnegative ridge weight. For a two-output full-QR procedure matching our motivating video analysis, cumulative edit rank similarly changes from two to the ambient dimension four. Conversely, the complete cumulative metric-QR trajectory is affine-equivariant when its positive-definite metric, probe, regularizer, and tie-breaking are transported consistently; exact covariance is one corollary, not a canonical semantic metric. In a known-rank finite-sample Adam/QR calibration, identity mixing stops after one accepted update in all 20 large-sample runs, whereas each tested shear $a\in\{.5,.75,1,1.25,2\}$ accepts at least two updates in all 20 runs. Controlled reparameterizations of frozen V-JEPA2 features preserve rank-zero predictions yet alter later Euclidean trajectories under practical optimization. These visual contact experiments are stress tests, not estimates of contact dimension. Iterative erasure therefore returns a procedure-relative estimand jointly determined by representation geometry and the full measurement procedure, not a semantic dimension by itself.

Contrastive learning and Siamese embedding models have become the foundation of modern verification systems, where decisions are governed not by discrete classification boundaries, but by relational geometry in embedding space. However, existing adversarial attacks remain fundamentally classification-centric, overlooking the vulnerability of relational geometry. In this paper, we introduce a geometry-aware adversarial attack framework that reformulates attacks on contrastive systems as manifold-level relational corruption. Instead of targeting individual predictions, the proposed framework systematically distorts similarity organization within the embedding manifold by pushing positive pairs apart while simultaneously pulling negative pairs closer, ultimately collapsing and inverting pairwise similarity structure. To enable scalable deployment, we shift iterative online optimization into an offline adversarial geometry deformation prior learning stage and train a lightweight feed-forward generator that learns generalized geometry deformation patterns from the victim model. Once trained, the generator produces adversarial perturbations through a single forward pass without requiring online gradient computation, enabling real-time online attacks against similarity-based verification systems. Experimental results across multiple verification architectures demonstrate substantial degradation of verification performance together with severe manifold-level relational corruption. On the Markmatch verification system, the proposed attack reduces accuracy from 95.4% to 38.6% while completely reversing the positive-negative similarity structure.

Transformers have had a profound impact on the world of language processing and computer vision. As efforts to answer the million-dollar question of ``How does a Transformer learn?" have been increasing, existing interpretability studies primarily analyze representations at isolated layers or the network as a whole, while the developmental evolution of individual representations and its manifolds across transformer layers remains underexplored. With this work, we aim at providing a comprehensive analysis of the evolution of representations as the representation point cloud transforms across the layers; thereby attempting to isolate layers or establish a trend which comes closer to justifying how and when raw input representations evolve into task-relevant feature representations. Thus, Transformer Geometry Observatory-TGO-IV introduces a topological framework for analysing the evolution of Transformer representations through the lens of Persistent Homology. Rather than studying local geometric properties alone, TGO-IV constructs Vietoris--Rips simplicial complexes from token-level representation point clouds and investigates the evolution of their persistent topological signatures across Transformer layers. The proposed framework comprises complementary topological observatories including Persistence Diagrams, Barcode Diagrams, Betti Curves, Persistence Landscapes, Bottleneck Distance, and Wasserstein Distance, enabling a comprehensive analysis of how the global topology of representation point clouds develops throughout the forward pass.