arXiv subjects

Kaizhen Tan

Publications and source records attributed to Kaizhen Tan.

At least 19 recordsLinked to original sources

arXiv2026

You Cannot Photograph the Same Street Twice: Reliability Limits in Vision-Language Measurement of Urban Change

Vision-language models (VLMs) are increasingly used to measure urban change from repeated street-level imagery, and the results are often mapped for individual sample points. Such maps assume that a perception score changes only if the street does. We test this assumption with repeated Google Street View captures of unchanged streets in five US cities and with controlled experiments that hold the photograph, scene or camera fixed. Re-photographing an unchanged street shifts its perception score by two-thirds of the average difference between two streets in the same city. Much of this shift arises in the scoring and averages out over question orders; the rest comes from the photographs and is not predicted by image statistics of weather and light. A single image measures a place moderately well, but the change at one location barely separates redeveloped from unchanged streets. All models tested score degraded images as worse-looking streets, and in crowdsourced imagery camera differences cause false reports of physical change unless both images are rendered through a common virtual camera. After aggregation, redeveloped streets are scored as wealthier and better maintained, with more enclosure and less greenery. Streets photographed in the same capture campaign share part of their error, which averaging within a district does not remove. For urban research and planning, VLM perception scores can support comparisons between groups of streets, such as redeveloped and unchanged streets photographed in the same campaigns, but not the identification of individual streets that improved or declined.

cs.CV↗

arXiv2026

Confidence under Visual Token Pruning: Removed Evidence and Risk-Controlled Token Budgets for MLLMs

Visual token pruning speeds up multimodal large language models (MLLMs) by keeping a small subset of the visual tokens, and pruning methods are compared by the accuracy they retain. We study what pruning does to the confidence of these models, across common selectors, several MLLMs, and different output formats. Pruning errors concentrate on questions whose evidence the selector removed, and confidence does not register the removal. When the queried object loses all of its tokens, accuracy on these questions drops from 59% to 17%, while confidence stays at the unpruned level. Returning a few object tokens to the kept set recovers most of the lost accuracy. Selectors that keep the most attended tokens remove such evidence most often and produce confident errors, which temperature scaling cannot re-rank. Selectors that avoid keeping redundant tokens stay close to the calibration of the unpruned model. We then use the confidence of the pruned model to set a per-question token budget. The model answers with few tokens first and again with all tokens when its confidence is low. Conformal risk control sets the threshold to bound the expected deviation from the unpruned model. With coverage-based selection, this cascade needs about a third of the prefill tokens of the unpruned model, while with FastV it needs more than the unpruned model. The savings come mainly from how well the confidence ranks the answers that differ from the unpruned ones.

cs.CV↗

arXiv2026

How Much Can a LoRA Adapter Memorize? Measuring Adapter Capacity in Bits

LoRA adapters are often shared on their own, and the amount of information they can hold about their training data bounds what sharing them can reveal. Following Morris et al. (2026), we measure this amount in bits by training adapters on frozen language models to memorize random token sequences. LoRA adapters store 2 to 3.4 bits per trainable parameter, less than full fine-tuning of the same model. The number of parameters alone does not set this amount: at equal size, adapters on MLP layers store more than adapters on attention layers, and a randomly initialized frozen model supports as much storage as a pretrained one once the scale of its output logits can be trained. We then train the same adapter on a task with supervised fine-tuning (SFT) and with GRPO. At the same accuracy, SFT stores about three times as much information specific to its training examples, and it stores most of the bits of secrets planted in the training data, which GRPO does not store because it rarely samples them.

cs.LG↗

arXiv2026

Vision Models Predict Urban Scene Appraisal with Limited Neural Alignment

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.

cs.CV↗

arXiv2026

Teaching Vision-Language Models to Use the Scale They Are Given: Label-Free Equivariance Training for Metric Physical Reasoning

Metric questions about video, such as the speed of a moving object, require a vision-language model to convert visual measurements into physical units using a real-world reference supplied in the prompt. We find that current models use this reference only partially. When every world-space quantity in the prompt is multiplied by a common factor, the prompt still describes the same video and the correct answer changes by exactly that factor, but the predictions of eight models change by less, and their accuracy stays concentrated near the scale that the depicted objects usually have. Asked the same physics in a scale-free form, the two models we test recover the closed-form scaling laws on most items, which indicates that the deficit lies in metric grounding and not in knowledge of the physical mechanism. Because the scaling relation is exact, it can serve as supervision without metric annotations. Equivariance Self-Distillation (EquiSD) projects a model's own prediction onto the functions that satisfy this relation and fine-tunes the model on the resulting targets, with one query per training question and no ground truth. Trained on synthetic video only, EquiSD brings a 3B model close to the exact relation on held-out simulated videos, also at scales not seen in training, and improves its accuracy across scales. Without adaptation, it also improves accuracy across scales on the QuantiPhy benchmark, where its gain reaches 93% of that obtained by supervision with exact simulator answers.

cs.CV↗

arXiv2026

Seeing the City or Recognizing the Place? What Street-View Imagery Adds Beyond Existing Urban Data in VLM-Based Urban Sensing

Street-view imagery is increasingly analysed with vision-language models (VLMs) to infer urban attributes, but predictive accuracy alone does not show how much a photograph contributes beyond data already available for the same place. Using three VLMs, we compare image-based predictions with existing urban data for seven attributes drawn from five public resources. Each urban unit is evaluated with imagery, with location or text context, and against non-image predictions from nearby observations or public records. We also replace images and add conflicting records to test which source the predictions follow. Nearby observations or public records matched or exceeded image-only predictions for road damage, curb ramps, population, and house price. Images were more informative for building type, building function, and the floor count of low-rise buildings. For floor count, the advantage of images over nearby OpenStreetMap labels increased by 5.7 percentage points per doubling of distance to the nearest labelled building and declined for tall buildings whose rooflines often fell outside the frame. When images and records disagreed, predictions usually moved towards the supplied record. Released OpenFACADES floor annotations, generated with OpenStreetMap floor values as input, showed the same dependence: their agreement with the reference increased with building height, whereas that of image-only reruns decreased. The value of street-view imagery therefore depends on whether an attribute is visible and how well the place is already covered by existing data. Because machine-derived labels are often reused as references, these comparisons also bear on how urban datasets are documented and evaluated.

cs.CV↗

arXiv2026

Early Learning Shapes Later Directions Of Representation Change In Continual Learning

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.

cs.LG↗

arXiv2026

GeoFidelity-Bench: Evaluating Block-Conditioned Geographic Fidelity in Text-to-Image Street-View Generation

Text-to-image models can generate plausible street scenes from prompts such as "a street in New York City," but it remains unclear whether these images resemble the requested place or show a generic city appearance. Commonly used evaluation metrics mainly measure image quality or text alignment and do not directly assess geographic fidelity. We introduce GeoFidelity-Bench, a benchmark for evaluating whether generated street-view images match a target location at the level of named street blocks. The benchmark contains 112 street blocks from 25 cities across six continents, each paired with a panel of real Mapillary images and geographic metadata. From this metadata we build prompts with the city only, with street and neighborhood names, and with these names plus raw GPS coordinates. Across six open-weight text-to-image models, street and neighborhood names improve similarity to the target location, while raw GPS coordinates add only a small and uneven gain. Control prompts that swap in the names of other blocks while keeping the city fixed reduce retrieval accuracy, with larger decreases when the neighborhood name or both names are replaced. Generated images also distinguish their target blocks from local alternatives less reliably than held-out real images. CLIP alignment between generated images and their prompts is nearly uninformative about whether the images achieve block-level geographic fidelity. GeoFidelity-Bench provides a controlled testbed for studying location-conditioned generation and for measuring progress beyond generic city-level plausibility.

cs.CV↗

arXiv2026

Beyond Multimodal Alignment: Shared Physical Representations Across Sensors and Action Orders

Multimodal world models are often evaluated by whether different sensors produce similar representations. However, similar representations do not necessarily imply that the models make the same physical predictions, or that those representations can be reused when actions are combined in a new order. We study both questions through the physical responses predicted by a model. We first use the Cluster Haptic dataset to ask whether audio and acceleration can independently recover the behavior of the same surface from different observations. Predictions from the two sensors are substantially closer for the same surface than for different surfaces, with a $4.5\times$ gap on average, while both also outperform an average-surface prediction. We then show that this agreement alone does not determine how familiar actions should compose. In a controlled elastoplastic system, shared step dynamics fit observed programs less accurately than a whole-program predictor but generalize better to unseen action orders, with the ranking reversing on both held-out transitions across three independent initializations. Fusing free-decay and hysteresis observations further improves prediction, with diagonal Gaussian beliefs yielding the lowest errors. Together, these results distinguish cross-sensor consistency, multimodal fusion, and generalization to new action orders as separate questions in evaluating multimodal physical representations.

cs.LG↗

arXiv2026

What Can Latent World Models Know? Physical Information in Multimodal Predictive Representations

A central premise of latent world models is that predicting the future encourages representations to internalize the physics of their environment. We ask which physical quantities are accessible in learned latent states, how this depends on training, and how those quantities relate to the model's predictions. We present PokeWorld, a simulated environment in which a robot finger pushes objects whose mass, drag, and contact stiffness vary across episodes while remaining visually identical. We first measure parameter recovery from raw observation sequences, then train matched action-conditioned world models. Prediction targets can strongly shape latent content. For example, contact stiffness becomes decodable when touch is predicted, while providing touch only as an input does not. Longer-horizon prediction improves estimation of position and velocity from learned representations relative to single-step prediction. Drag is recoverable from raw observations, but has weak linear readout from the learned latent states. Yet the models' glide forecasts depend systematically on drag. Experiments on RH20T reproduce the same input-target dependence on real multimodal robot data. Together, these results show that observations, prediction targets, and prediction horizons shape both which physical quantities are accessible in latent states and how they affect future predictions.

cs.LG↗

arXiv2026

PopNavShift: Stress-Testing Social Navigation under Behavioral Population Shift

Social-navigation algorithms are often evaluated under a fixed pedestrian-behavior distribution, despite substantial variation in pedestrian responses to robots across individuals and social contexts. We introduce PopNavShift, a matched simulation framework for stress-testing social-navigation strategies under pedestrian population shifts. PopNavShift constructs population-conditioned pedestrian motion profiles by prompting Gemini 3.7 Flash with 600 synthetic persona records from MatrAIx Persona 1M and deterministically mapping the responses into bounded motion parameters. It then compares three representative navigation strategies, reactive avoidance, early yielding, and reciprocal collision avoidance, across eight population conditions and 7,488 matched robot runs. In a matched intervention on the same 202 personas, changing only time pressure reverses 8.6% of controller rankings based on robot travel time, but 22.4% based on mean pedestrian delay and 23.9% based on worst-decile delay. Across population conditions, this sensitivity is greater for pedestrian burden than for robot travel time and increases in spatially constrained settings; the same qualitative pattern persists under a second pedestrian dynamics model. These findings support evaluating navigation strategies across behavioral populations using both robot performance and pedestrian burden.

cs.RO↗

arXiv2026

CacheWeaver: Cache-Aware Evidence Ordering for Efficient Grounded RAG Inference

Retrieval-Augmented Generation (RAG) improves factual grounding, but it also lengthens prompts and raises prefill cost. Prefix caching in serving engines such as vLLM reduces this cost only when requests share the same token prefix. In grounded generation, however, adjacent queries may retrieve overlapping evidence in different orders, so set overlap does not become reusable prefix overlap. We present CacheWeaver, a lightweight prompt-layer method for cache-aware evidence ordering. The method keeps a prefix tree over recently served evidence sequences and uses a greedy walk to place the most reusable prefix first, while leaving the serving engine and retrieved evidence set unchanged. Across three vLLM configurations, the method lowers median time-to-first-token (TTFT) by about 20-33 percent relative to retrieval-order prefix caching, without hurting answer quality in our QA tests. The greedy policy reaches 97.5 percent of the median TTFT gain from oracle ordering, indicating that most reusable prefix locality can be recovered by a simple scheduling layer between retrieval and inference.

cs.CL↗

arXiv2026

UrbanVGGT: Scalable Sidewalk Width Estimation from Street View Images

Sidewalk width is an important indicator of pedestrian accessibility, comfort, and network quality, yet large-scale width data remain scarce in most cities. Existing approaches typically rely on costly field surveys, high-resolution overhead imagery, or simplified geometric assumptions that limit scalability or introduce systematic error. To address this gap, we present UrbanVGGT, a measurement pipeline for estimating metric sidewalk width from a single street-view image. The method combines semantic segmentation, feed-forward 3D reconstruction, adaptive ground-plane fitting, camera-height-based scale calibration, and directional width measurement on the recovered plane. On a ground-truth benchmark from Washington, D.C., UrbanVGGT achieves a mean absolute error of 0.252 m, with 95.5% of estimates within 0.50 m of the reference width. Ablation experiments show that metric scale calibration is the most critical component, and controlled comparisons with alternative geometry backbones support the effectiveness of the overall design. As a feasibility demonstration, we further apply the pipeline to three cities and generate SV-SideWidth, a prototype sidewalk-width dataset covering 527 OpenStreetMap street segments. The results indicate that street-view imagery can support scalable generation of candidate sidewalk-width attributes, while broader cross-city validation and local ground-truth auditing remain necessary before deployment as authoritative planning data.

cs.CV↗

arXiv2026

Renaming or Tightness: Enforcing Disjunctive Information Flow Policies

A disjunctive policy allows a value to depend on at most one of two secrets and never on both: an analyst may consult one client's file or the other's, a share of a split secret may be released but not its sibling. Such policies are not lattice-shaped, and Hunt and Sands introduced the quantale of information to give them a semantics, leaving the enforcement layer open. We build the flow-sensitive type system family that the quantale calls for, and show that the object which makes such families useful, the universal type object from which every member specialises, splits in two, with a consequence for enforcement. Over the free commutative quantale on the program variables the whole mechanism survives for every policy: monotone renaming, canonical derivations, principal typings, internal completeness. Over the free object with idempotent generators the certified bound is strictly more precise and still sound, because it records that two reads of one source honour one disjunct. The gap cannot be closed from inside the independent-attribute family: no mechanism of that shape whose labelling maps support monotone renaming certifies a bound more precise than the first, and for principal typings under generator-exact homomorphic specialisation the two coincide. Under the ethical-wall and secret-sharing labels the second read of a disjunctive source therefore drives every such certificate to no guarantee, and programs that satisfy the policy are rejected. The literal transcription of the lattice-era object is no escape either: it is a further quotient that loses branch disjunction. Precision is recovered by deferring specialisation to the judgement level, and the resulting read-out map is the least sound join-preserving one.

cs.CR↗

arXiv2026

Collusion with Competitive Marginals: Price-Level Audits Are Blind by Construction

Empirical work on algorithmic collusion asks one question of the data: are prices supracompetitive? We show this can be answered "no" by a conspiracy that is nonetheless profitable. Consider bidding agents that couple only through the joint distribution of their unexplained bid components, leaving every agent's own bid law exactly at the competitive law. Any test whose input is a single agent's price or bid history then has power exactly equal to its false-positive rate, for every coupling strength up to comonotonicity. The published detection methodology is therefore blind to this conduct by construction rather than underpowered, and no sample size repairs it. Three empirical results follow. First, the mechanism appears in real language-model agents: twenty models from nineteen independent developers, three deployment prompts each, show residual correlation of $+0.053$ between two deployments of one model against $+0.0001$ across models, with a 95% interval clustered by developer of $[0.030, 0.078]$, under an auditor that sees every order feature and is fitted out of sample. Second, the coupling falls monotonically as sampling temperature rises ($p=0.002$), turning a deployment parameter into a candidate mitigation. Third, on 24 days of Ethereum block-building auction data covering 77,684 bids from 39 bidders, the honest population of bidder pairs is itself so dependent that a screen held at a 5% false-positive rate must sit above a floor of $+0.50$ to $+0.81$, which is 20 to 32 times the family-wise sampling threshold and does not fall as the audit window grows. Since lawful multi-identity operation and conspiracy are behaviourally indistinguishable here, the tractable regulatory target is not detection but counting: resolving 40 bidding identities into 23 operators raises the Herfindahl index by 247.5%, and adding behavioural clusters from public bid streams reaches 324.5%.

cs.GT↗

arXiv2026

Automorphism-Induced Non-Canonicity in Top-k Explanations of Graph Neural Networks

A gradient-based GNN explainer given a molecule with two chemically equivalent nitro groups assigns them attribution scores that are equal to the last bit. It cannot do otherwise: message passing is exactly permutation equivariant, so any automorphism of the input leaves every attribution invariant. Yet the standard report, the top-k edges, names one of the two, and which one is settled by the order of an array. We show this is a structural obstruction rather than an implementation slip. When no minimal valid explanation is fixed by the input's automorphism group, no rule can be single-valued, minimal and symmetry-respecting at once. For the exact-k reports used in practice we give a parameter-free criterion, mechanised in Lean 4 with no axiom dependencies, that decides from the graph alone whether every score-optimal report of that size must split an orbit. Across 21298 instance-budget decisions the criterion agrees with a mechanical model-equivalence check without exception, and no severing case we found admitted a neutral alternative. The obstruction is common. Nontrivial automorphisms occur in 93.4% of Mutagenicity, the dataset the seminal explainability papers use, so the measure-zero dismissal of symmetric inputs, sound on the continuous domains it was made for, collapses here. At the sparsity budget those papers report, 24.0% of molecules with two interchangeable nitro groups (6 of 25) surface exactly one of them, every one arbitrary under mechanical verification. A model's blindness also manufactures symmetry: every MUTAG molecule contains atoms chemistry separates and the network provably cannot, and a matched control shows the resolution is set by what the model reads rather than how it is parameterised. Reporting orbits removes the arbitrariness at 0.11 ms and 0.43 extra edges per graph.

cs.LG↗

arXiv2026

CREG: Compass Relational Evidence Graph for Characterizing Directional Structure in VLM Spatial-Reasoning Attribution

Standard attribution heatmaps show where a vision-language model (VLM) focuses, but they do not reveal whether the recovered evidence is organized by the queried spatial relation or merely reflects image layout. To address this problem, we introduce CREG (Compass Relational Evidence Graph), a training-free diagnostic framework that converts token-level attribution into a reference-centered compass distribution and measures its directional alignment. CREG provides a shared directional readout across attribution methods and makes comparison with geometric controls explicit. Across three spatial-relation benchmarks, box-only geometry achieves Direction Alignment Error 28.4 to 34.4 degrees lower than the best current model-based attribution method on each dataset, leaving a substantial gap between attribution structure and simple target localization. To examine this gap, we apply a diagnostic battery including target intervention, reference-center randomization, and variance partition. Taken together, the results suggest that the directional structure recoverable from current attribution methods is limited and often mixed with image layout. We further find that higher task accuracy does not reliably coincide with better directional attribution: small-scale LoRA training and newer model generations can improve task accuracy while leaving Direction Alignment Error unchanged or worse. These findings characterize what current attribution methods reveal rather than the model's internal spatial representation. CREG provides a controlled protocol for testing whether improvements in spatial reasoning are accompanied by more directionally organized evidence.

cs.CV↗

arXiv2026

Decoding Tourist Perception in Historic Urban Quarters with Multimodal Social Media Data: An AI-Based Framework and Evidence from Shanghai

Historic urban quarters are increasingly shaped by tourism and lifestyle consumption, yet planners often lack scalable evidence on what visitors notice, prefer, and criticize in these environments. This study proposes an AI-based, multimodal framework to decode tourist perception by combining visual attention, color-based aesthetic representation, and multidimensional satisfaction. We collect geotagged photos and review texts from a major Chinese platform and assemble a street view image set as a baseline for comparison across 12 historic urban quarters in Shanghai. We train a semantic segmentation model to quantify foregrounded visual elements in tourist-shared imagery, extract and compare color palettes between social media photos and street views, and apply a multi-task sentiment classifier to assess satisfaction across four experience dimensions that correspond to activity, physical setting, supporting services, and commercial offerings. Results show that tourist photos systematically foreground key streetscape elements and that the color composition represented on social media can differ from on-site street views, indicating a perception-reality gap that varies by quarter. The framework offers an interpretable and transferable approach to diagnose such gaps and to inform heritage management and visitor-oriented urban design.

cs.AI↗