arXiv:2607.16171v1 Announce Type: cross
Abstract: We disprove the conjecture that every globally asymptotically stable homogeneous polynomial vector field admits a homogeneous polynomial Lyapunov function. The counterexample is a planar homogeneous cubic polynomial vector field with integer coefficients. It admits no positive definite homogeneous polynomial with nonpositive Lie derivative and, more strongly, no real-analytic Lyapunov function even locally. Nevertheless, it has an explicit degree-two homogeneous Lyapunov function that is radially unbounded, continuously differentiable everywhere, and smooth away from the origin. We also provide a machine-checked Lean 4 formalization of the main result.
Science Journals
arXiv:2607.15587v1 Announce Type: new
Abstract: Continual learning studies how deployed language models can continually acquire new tasks without expensive retraining from scratch. Existing methods, whether rehearsal-based (replaying stored past data) or rehearsal-free (regularising or isolating parameters), overwhelmingly target one objective: preventing catastrophic forgetting. Forward transfer, the past helping the future, has meanwhile been pursued almost exclusively through parameter reuse, with no explicit account of when transfer should be expected at all. We begin one step earlier: before designing a transfer mechanism, we ask when transfer should exist at all. We answer with a framework of three measurable conditions: the target task must leave room for improvement beyond its own limited supervision, transferable information must survive continued optimisation, and replay must come from compatible previous tasks. We instantiate this view as Transfer-Selective Replay (TSR), which selects replay data predicted to benefit the incoming task rather than replaying past examples indiscriminately. Selection is guided by a zero-training task signature, while distillation preserves stability on previous tasks. Under the standard continual learning protocol in the low-budget regime, TSR consistently improves forward transfer while maintaining stability, outperforming existing replay baselines across heterogeneous and homogeneous task streams. More broadly, the results argue for treating transfer as a first-class objective of continual learning, to be understood before it is engineered.
arXiv:2507.15787v3 Announce Type: replace
Abstract: Modelling physical systems with partial differential equations (PDEs) is central to science and engineering, yet in most real applications the PDE model is incomplete: relationships such as constitutive or thermal laws are unknown. Existing surrogate approaches close this gap by learning the PDE solution from data, but remain tied to a specific configuration (geometry, boundary conditions, discretisation) and recover the solution rather than the missing physics itself.
We introduce FEML, an end-to-end differentiable framework that couples the known PDE with a machine-learned operator for the missing physics. Embedding the PDE solver into training lets this operator be learned directly from the PDE solution, even when its own output cannot be measured - for example, stress in constitutive laws. Because the operator is independent of the system configuration, a law learned in one setting transfers zero-shot to new geometries, boundary conditions, and discretisations, and can be inspected by domain specialists. FEML represents the operator with structure-preserving operator networks (SPONs), which retain key continuous properties at the discrete level.
We demonstrate FEML across solid mechanics and thermal transport. From synthetic data we progressively discover an elastoplastic law - the nonlinear elastic response, then the plastic hardening law - and compose them into a foundation constitutive model that transfers zero-shot to a 3D torsion problem. Moving to real data, we learn coupled plastic-hardening and ductile-damage laws from a benchmark shear-coupon test, reproducing the measured response, including post-peak softening, to within the experimental scatter. Finally, we recover a temperature-dependent conductivity from transient heat-flow data and apply symbolic regression to the learned operator to extract a closed-form law matching the ground truth.
arXiv:2507.22854v3 Announce Type: replace
Abstract: We propose novel classical and quantum online algorithms for learning finite- and infinite-horizon Markov Decision Processes (MDPs). Our algorithms are based on a hybrid online-offline reinforcement learning model wherein the agent can, from time to time, freely interact with the environment in a generative sampling fashion, i.e., by having access to a "simulator". By employing known classical and new quantum algorithms for approximating optimal policies under a generative model within our learning algorithms, we show that it is possible to avoid several paradigms from RL like "optimism in the face of uncertainty" and "posterior sampling" and instead compute and use optimal policies directly, which yields better regret bounds compared to previous works. Our quantum algorithms obtain regret bounds which only a $\operatorname{poly}\log{T}$ dependence on the number of time steps $T$, thus breaking the $O(\sqrt{T})$ classical barrier. Our infinite-horizon discounted regret bound is brand new, while in the finite- and infinite-horizon undiscounted settings, our results match the time dependence of some prior quantum works, but with improved dependence on other parameters like state space size $S$ and action space size $A$.
arXiv:2508.00042v2 Announce Type: replace
Abstract: Machine learning models deployed in non-stationary environments degrade silently, since as the input distribution drifts their accuracy decays without an error signal and without labels to reveal it. Sustaining reliable AI therefore requires a concept-drift detector that acts as an external observer of the deployed model, monitoring it using unlabeled operational data alone, so that an MLOps actuator triggers retraining and redeployment only when it is warranted. This paper contributes two concept drift detectors, namely Confidence-Filtered Pseudo-Label Transfer (CFPT) and TabAutoDrift, which combine representation learning with statistical testing to compute an expected utility score that signals whether a deployed model should be retrained, without requiring ground-truth labels after deployment. The detectors are evaluated on two emerging, label-scarce wireless application domains in which post deployment ground truth is effectively unavailable, namely outdoor fingerprinting-based localization and link-anomaly detection. They outperform the classical detectors ADWIN, DDM, and CUSUM, attaining a drift-detection F1-score between 0.88 and 0.94 in the fingerprinting use case and between 0.80 and 1.00 in the link-anomaly use case, up to 0.24 higher than the strongest classical detector. Interpreted as reliability decisions, this precision indicates that the proposed detectors signal retraining more dependably.
arXiv:2508.00521v3 Announce Type: replace
Abstract: Accurate measurement of surface plasmon polariton (SPP) dispersion remains challenging, as conventional angle-resolved techniques are limited by surface quality, diffraction artifacts, and geometry-induced band folding. Here, we show that SPP dispersion can be reconstructed from transmission spectra of plasmonic gratings with subwavelength apertures acting as Fabry-P\'erot (FP) cavities. The approach harnesses modal hybridization between localized FP modes and SPPs, resolved using non-Hermitian eigenmode decomposition and validated by finite-difference time-domain simulations. {\omega}-k dispersion mapping is achieved by varying the grating periodicity, with each structure probing a distinct in-plane momentum state. Geometry- and material-dependent corrections for aperture-induced leakage and dispersive phase shifts yield reconstructed relations in close agreement with eigenmode calculations across non-dispersive, Drude, and Drude-Lorentz models as well as experimental optical datasets spanning metals, oxides, and nitrides. The method is material-agnostic and requires no momentum-resolved instrumentation. Sensitivity to fabrication-induced wall roughness is also assessed: FP resonance positions remain spectrally stable with no measurable linewidth broadening across the explored perturbation range, and the modal field topology is largely preserved throughout. However, transmitted amplitude decreases monotonically owing to enhanced ohmic absorption at the perturbed boundaries.
arXiv:2506.19657v3 Announce Type: replace
Abstract: The Circular Economy framework emphasizes sustainability by reducing resource consumption and waste through the reuse of components and materials. This paper presents ReLink, a computational framework for the circular design of planar linkage mechanisms using available standard parts. Unlike most mechanism design methods, which assume the ability to create custom parts and infinite part availability, ReLink prioritizes the reuse of discrete, standardized components, thus minimizing the need for new parts. The framework consists of two main components: design generation, where a generative design algorithm generates mechanisms from an inventory of available parts, and inverse design, which uses optimization methods to identify designs that match a user-defined trajectory curve. The paper also examines the trade-offs between kinematic performance and CO2 footprint when incorporating new parts. Challenges such as the combinatorial nature of the design problem and the enforcement of valid solutions are addressed. By combining sustainability principles with kinematic synthesis, ReLink lays the groundwork for further research into computational circular design to support the development of systems that integrate reused components into mechanical products.
arXiv:2607.15297v1 Announce Type: cross
Abstract: This paper proposes a large language model-enhanced multi-hop parallel image semantic communication (LLM-MHPSC) framework to mitigate distortion accumulation in multi-hop wireless image transmission. Unlike conventional single-hop semantic communication schemes, LLM-MHPSC deploys an extra residual compensation link at each hop to counteract accumulated distortions. To minimize additional bandwidth overhead, a coarse-to-fine residual compression scheme is designed by integrating a deep learning-based compressor with adaptive arithmetic coding (AAC). Furthermore, a large language model-based residual transmission optimizer (LLM-RTO) is developed to accurately estimate residual distributions and enable channel state and hop-aware rate adjustment, thereby improving residual compression efficiency under varying channel and hop conditions. An adaptive hop selection strategy is also proposed to activate the residual link on demand, striking a balance between transmission performance and computational cost. Experimental results show that LLM-MHPSC outperforms state-of-the-art semantic communication and traditional schemes, realizing robust image transmission with a marginal increase in bandwidth. This framework provides a flexible and effective solution for extending semantic communication to practical multi-hop application scenarios.
arXiv:2508.05321v4 Announce Type: replace
Abstract: Assume you encounter an inverse problem that shall be solved for a large number of data, but no ground-truth data is available. To emulate this, in this study we assume it is unknown how to solve the imaging problem of Computed Tomography. We introduce an unsupervised deep learning framework that leverages the inherent similarities between iterative reconstruction, Deep Image Prior (DIP), and unrolled optimization schemes. Our specific contribution is a training framework for amortized reconstruction: After training on a dataset without any image-domain ground truth, reconstruction of an unseen scan reduces to a single network forward pass. We demonstrate the feasibility of reconstructing images from measurement data by pure network inference, without additional gradient steps for unseen samples. Our method is evaluated on the two-dimensional 2DeteCT dataset. Within a controlled, geometry-matched benchmark, our reconstructions are competitive with, or better than, filtered back-projection, maximum-likelihood reconstruction, and a supervised network of identical architecture. Compared to a per-image DIP baseline, our method reaches similar quality while replacing the costly per-instance optimization with a single forward pass, yielding a speed-up of about four orders of magnitude. This makes it a promising candidate for time-critical imaging applications. Future work will address multi-dataset adaptability, counter-measures against over-smoothing, advanced uncertainty quantification, and further medical-imaging inverse problems.
arXiv:2508.09860v2 Announce Type: replace
Abstract: Human-aligned AI is a critical component of co-creativity, as it enables models to accurately interpret human intent and generate controllable outputs that align with design goals in collaborative content creation. This direction is especially relevant in procedural content generation via reinforcement learning (PCGRL), which is intended to serve as a tool for human designers. However, existing systems often fall short of exhibiting human-centered behavior, limiting the practical utility of AI-driven generation tools in real-world design workflows. In this paper, we propose VIPCGRL (Vision-Instruction PCGRL), a novel deep reinforcement learning framework that incorporates three modalities-text, level, and sketches-to extend control modality and enhance human-likeness. We introduce a shared embedding space trained via quadruple contrastive learning across modalities and human-AI styles, and align the policy using an auxiliary reward based on embedding similarity. Experimental results show that VIPCGRL outperforms existing baselines in human-likeness, as validated by both quantitative metrics and human evaluations. The code and dataset are available at https://github.com/bic4907/VIPCGRL.
arXiv:2508.10029v3 Announce Type: replace
Abstract: Safety-aligned large language models can still be manipulated through white-box interventions that modify their internal representations. We introduce Latent Fusion Jailbreak (LFJ), which works by pairing a harmful query with a structurally similar but benign counterpart, then interpolating their hidden states at carefully selected layers and token positions. Refusal-loss gradients determine exactly where to intervene, and we optimise layer-wise mixing coefficients using token-normalised compliance and refusal-suppression objectives. The edited prompt states propagate sequentially through the remaining transformer blocks. Across four safety benchmarks and five open-weight target models, LFJ reaches a macro-averaged attack success rate (ASR) of 94.13% under the white-box protocol we describe. Because LFJ directly accesses internal states, comparisons with prompt-only attacks serve as a descriptive reference rather than a matched evaluation. Dropping rejection sampling lowers ASR to 86.72%, whereas replacing the structured harmful-benign pairing with random pairing causes it to fall to 27.45%. We also design an LFJ-specific latent adversarial training procedure that, when the attack is re-optimised against the defended model, reduces ASR from 94.13% to 12.37%. This defence evaluation does not cover transfer to other attack types or preservation of benign utility.
arXiv:2508.11444v4 Announce Type: replace
Abstract: In a recent paper, Francis, Illickan, Jose and Rajendraprasad showed that every $n$-vertex plane graph $G$ has (under some natural restrictions) a vertex-partition into two sets $V_1$ and $V_2$ such that each $V_i$ is \emph{dominating} (every vertex of $G$ contains a vertex of $V_i$ in its closed neighbourhood) and \emph{face-hitting} (every face of $G$ is incident to a vertex of $V_i$). Their proof works by considering a supergraph $G'$ of $G$ that has certain properties, and among all such graphs, taking one that has the fewest edges. As such, their proof is not algorithmic. Their proof also relies on the 4-color theorem, for which a quadratic-time algorithm exists, but it would not be easy to implement.
In this paper, we give a new proof that every $n$-vertex plane graph $G$ has (under the same restrictions) a vertex-partition into two dominating face-hitting sets. Our proof is constructive, and requires nothing more complicated than splitting a graph into 2-connected components, finding an ear decomposition, and computing a perfect matching in a 3-regular plane graph. For all these problems, linear-time algorithms are known and so we can find the vertex-partition in linear time.
arXiv:2508.12620v2 Announce Type: replace
Abstract: Large language models (LLMs) have recently shown impressive results on diverse code-related tasks, benefiting from large-scale training and instruction tuning. However, studies reveal that their grasp of fundamental programming concepts, such as data flow and control flow, remains shallow, leading to fragile performance when code requires deeper reasoning. This limitation restricts the practical adoption of LLMs in real-world software development. To address this issue, this work introduces a counterfactual code augmentation framework combined with concept-aware tuning, designed to guide LLMs toward stronger conceptual understanding. Comprehensive evaluation across multiple models and benchmarks demonstrates the effectiveness of the proposed approach.
arXiv:2508.19198v3 Announce Type: replace
Abstract: In this paper we consider the numerical approximation of the incompressible surface Navier--Stokes equations on an evolving surface. For the discrete representation of the moving surface we use parametric finite elements of degree $\ell \geq 2$. In the semidiscrete continuous-in-time setting we are able to prove a stability estimate that mimics a corresponding result for the continuous problem. Some numerical results, including a convergence experiment, demonstrate the practicality and accuracy of the proposed method.
arXiv:2508.21516v3 Announce Type: replace
Abstract: A digital twin (DT) contains a set of virtual models of real systems and processes that are synchronized with their physical counterparts. In a setup in which contact with the physical world is maintained through sensors and actuators that are wirelessly connected to the DT's computing engine, DT alignment requires periodic status updates, while safety-critical messages and fault conditions call for low-latency anomaly reporting, creating a fundamental trade-off in how wireless resources are used. We present a medium access framework combining pull-based updates, centrally scheduled according to goal-oriented principles, with urgent push-based updates, for which transmission decisions are made directly by the sensors. This enables the system to quickly detect and recover from anomalies while maintaining DT alignment. We thus design a push-pull scheduler (PPS) that strikes a balance in the trade-off between DT alignment in normal conditions and anomaly reporting, optimizing resource usage and reducing DT drift by 20 - 30% with respect to state-of-the-art solutions while maintaining the same anomaly detection guarantees, or reducing worst-case anomaly detection times by 30 - 70% while meeting the same DT alignment conditions.
arXiv:2509.11974v2 Announce Type: replace
Abstract: Federated Learning (FL) enables collaborative model training among clients without centralising data, making it a widely adopted privacy-enhancing technology (PET). Despite its privacy benefits, FL remains vulnerable to orchestrator-driven privacy attacks. In this paper, we study an underexplored threat in which a dishonest orchestrator intentionally manipulates the aggregation process to induce targeted overfitting in local models of specific clients. Although prior work focuses on reducing information leakage during training, we emphasise early client-side detection of targeted overfitting, allowing clients to disengage before significant harm occurs. To this end, we propose three detection techniques -- label flipping, backdoor trigger injection, and model fingerprinting -- which enable clients to verify the integrity of the global aggregation. We evaluated our methods across multiple datasets and attack scenarios. In single-client attacks, all three methods detect orchestrator-induced overfitting within 1-2 training rounds with F1 scores up to 0.7. Scalability experiments further show that detection effectiveness is influenced by cohort composition and method parameters. These results demonstrate that client-side integrity testing can provide early, effective, and scalable detection, supporting safer deployment of FL systems.
arXiv:2509.13026v2 Announce Type: replace
Abstract: Strong functors and monads are ubiquitous in Computer Science. More recently, (strong) comonads have demonstrated their use in structuring context-dependent notions of computation. However, the dualisation of ``being strong'' property passed somehow unobserved so far. We argue that ``being costrong'' gives a different understanding of how functors can interact with monoidal structures. We shall see that the well-known correspondence between distributive laws $F T \to T F$ of an endofunctor $F$ over a monad $T$, on one hand, and extensions of $F$ to the Kleisli category of that monad, on the other hand, generalises from ordinary monads to graded ones. The gist here is to recognise that the costrength of a costrong functor is nothing but a ``graded'' distributive law. As such, ``being costrong'' is a structure that a functor may have. Examples of costrong functors with respect to different graded monads are provided, with emphasis to the cartesian case, and applications to optics and coalgebras are given.
arXiv:2509.13186v2 Announce Type: replace
Abstract: Phishers achieve large-scale attacks by using ready-to-deploy phishing websites (phishing kits) to rapidly launch campaigns that leverage specific data exfiltration, evasion, or mimicry techniques. In contrast, researchers and defenders continue to rely on manual analysis to identify features for kit fingerprinting. In this paper, we examine the link between a page's client-side behavior and the underlying phishing kit used, enabling automated aggregation of phishing pages. Our key insight is that client-side techniques make heavy use of browser APIs, which, in turn, differentiate underlying kits based on their feature sets. Using an instrumented browser and a URL fuzzing utility, we collected traces from 1,328,917 pages and recovered kit archives for 4,180 pages between August 2023 and January 2025. For the labeled subset, we find that clustering based on the set of browser APIs executed yields 98% accuracy in grouping them by the underlying kit. We also find that 434,495 phishing pages execute enough browser APIs to cluster into 9,306 clusters, compressing multi-lingual phishing pages across various domains into a single cluster. Our findings show that analysts and researchers can leverage the complexity of client-side phishing code to track phishers' kit deployments in the wild.
arXiv:2510.00182v2 Announce Type: replace
Abstract: While we know that large language models (LLMs) can solve some planning problems, we do not understand the extent of these capabilities for robotics. One promising direction is to integrate the semantic knowledge of LLMs with the formal reasoning of task and motion planning (TAMP). However, designing such systems is complicated by the myriad of choices for how to integrate LLMs within TAMP. We develop 16 algorithms that use LLMs to substitute key TAMP components. Our zero-shot experiments across 13750 evaluations and three domains reveal that LLM-based planners exhibit lower success rates and higher planning times than engineered systems. Providing geometric details increases the number of task-planning errors compared to pure PDDL descriptions, and (faster) direct LLM variants outperform (slower) reasoning variants in most cases. Code and results are available at https://github.com/jorge-a-mendez/llm-pddlstream.
arXiv:2510.01894v3 Announce Type: replace
Abstract: Many natural dynamic processes -- such as in vivo cellular differentiation or disease progression -- can only be observed through the lens of static sample snapshots. While challenging, reconstructing their temporal evolution to decipher underlying dynamic properties is of major interest to scientific research. Existing approaches enable data transport along a temporal axis but are poorly scalable in high dimension and require restrictive assumptions to be met. To address these issues, we propose Multi-Marginal temporal Schr\"odinger Bridge Matching (MMtSBM) from unpaired data, extending the theoretical guarantees and empirical efficiency of Diffusion Schr\"odinger Bridge Matching (arXiv:2303.16852) by deriving the Iterative Markovian Fitting algorithm to multiple marginals in a novel factorized fashion. Experiments show that MMtSBM retains theoretical properties on toy examples, achieves state-of-the-art performance on real-world datasets such as transcriptomic trajectory inference in 100 dimensions, and, for the first time, recovers couplings and dynamics in very high-dimensional image settings. Our work establishes multi-marginal Schr\"odinger bridges as a practical and principled approach for recovering hidden dynamics from static data.
arXiv:2510.03504v3 Announce Type: replace
Abstract: Connectivity is crucial in many multi-robot applications, yet balancing connectivity maintenance and fleet traversability in obstacle-rich environments remains challenging. Reactive controllers based on control barrier functions can preserve connectivity when it is initially satisfied, but often struggle with deadlocks in cluttered environments. We propose a real-time B\'ezier-based constrained motion planning algorithm, namely MPC--CLF--CBF, that produces trajectories and control inputs concurrently, subject to high-order control barrier function and control Lyapunov function constraints. Our motion planner supports connectivity-aware navigation in cluttered workspaces and recovers connectivity from initially disconnected configurations and after temporary obstacle-induced separation; it also provides analytic continuous-time derivatives, facilitating its application to agile differentially flat systems such as quadrotors. In simulations with $4$--$12$ robots, it maintains $95.8$--$100\%$ graph-connected time at $20\%$ obstacle density, compared with $48.9$--$61.3\%$ for MPC--CBF, with no observed collisions. We further validate the planner in a physical experiment with $8$ Crazyflie nano-quadrotors.
arXiv:2510.05509v3 Announce Type: replace
Abstract: Diffusion models are powerful deep generative models, but unlike classical models, they lack an explicit low-dimensional latent space that parameterizes the data manifold. This absence makes it difficult to perform manifold-aware operations, such as geometrically faithful interpolation or conditional guidance that respects the learned manifold. We propose a training-free Riemannian metric on the noise space, derived from the Jacobian of the score function. The key insight is that the spectral structure of this Jacobian separates tangent and normal directions of the data manifold; our metric leverages this separation to encourage paths to stay tangential to the manifold rather than drift toward high-density regions. To validate that our metric faithfully captures the manifold geometry, we examine it from two complementary angles. First, geodesics under our metric yield perceptually more natural interpolations than existing methods on synthetic, image, and video frame datasets. Second, the tangent-normal decomposition induced by our metric prevents classifier-free guidance from deviating off the manifold, improving generation quality while preserving text-image alignment.
arXiv:2510.05750v2 Announce Type: replace
Abstract: Graph neural networks (GNNs) have achieved remarkable success in node classification. Building on this progress, heterogeneous graph neural networks (HGNNs) integrate relation types and node and edge semantics to leverage heterogeneous information. Causal analysis for HGNNs is advancing rapidly, aiming to separate genuine causal effects from spurious correlations. However, whether HGNNs are intrinsically effective for node classification remains underexamined, and most studies implicitly assume rather than establish this effectiveness. In this work, we examine HGNNs for node classification from two perspectives: model architecture and heterogeneous information. We conduct a systematic reproduction across 21 datasets and 20 baselines, complemented by comprehensive hyperparameter retuning. To further disentangle the source of performance gains, we develop a causal mediation analysis framework that treats the introduction of heterogeneous relation information as the treatment, candidate structural properties as mediators, and node classification performance as the outcome. This framework first screens candidate mediators according to their treatment-induced changes and their associations with performance improvement, and then decomposes the total effect into mediated and direct effects. Our results lead to two conclusions. First, model architecture and complexity have no causal effect on node classification performance. Second, heterogeneous information exerts a positive causal effect primarily through increasing homophily and local-global distribution discrepancy, which makes node classes more distinguishable. The implementation is publicly available at https://github.com/YXNTU/CausalHGNN.
arXiv:2607.15949v1 Announce Type: new
Abstract: We propose a stochastic behavioral modeling framework, termed Gaussian behaviors, which augments a deterministic linear time-invariant (LTI) behavior with a Gaussian noise component. We show that this notion is a tractable subclass of stochastic behaviors and encompasses classical parametric stochastic LTI state-space system models as special cases. Analogously to deterministic LTI behaviors, the framework enables simple and tractable stochastic data-driven control methods. To this end, we obtain a method for prediction by conditioning the Gaussian behavior on the known part of the trajectory, which is identified directly from the sample covariance of trajectory data. Building on this method, we develop predictive control formulations that optimize over feedforward or disturbance affine feedback policies. The resulting formulations are shown to be convex. We further derive a finite-sample confidence bound on the prediction accounting for both aleatoric and epistemic uncertainty, and incorporate it into a robust control method, for which a tractable convex upper bound is obtained. Within this framework, subspace predictive control is recovered when only the mean prediction is used, while data-enabled predictive control is shown to account for the prediction uncertainty in an optimistic fashion. Numerical case studies illustrate the benefits of the proposed methods.
arXiv:2607.15951v1 Announce Type: new
Abstract: We present the first implementation of a 3D Gaussian renderer on an Intelligence Processing Unit (IPU), comprising 1,472 independent tiles with only on-chip SRAM; constraints that approximate properties of efficient sensor-processor architectures. Our input scenes are 3D Gaussian maps from real-world sequences. Each tile 'owns' a screen-space region of the framebuffer; Gaussian primitives are routed to destination tiles via Manhattan-distance hops on a north-east-west-south (NEWS) grid, then distributed to overlapping neighbours in an expanding tree pattern. Computation follows the IPU's Bulk Synchronous Parallel (BSP) model, with inter-tile communication defined at compile time. We show this hardware allows us to exploit spatial and temporal locality by enabling local data transfer between cores. We evaluate the bottlenecks in this SRAM-only implementation: inter-tile bandwidth, per-tile SRAM capacity, and workload imbalance from non-uniform Gaussian density. We analyse how these constraints affect performance and render quality. This exploration raises broader questions for conventional GPUs and 3D representations, suggesting that direct inter-SM (streaming multiprocessor) communication might offer ways to reduce DRAM access in GPU kernels. We discuss these implications for the future of on-sensor and DRAM-free architectures. Project page: https://nmjfry.github.io/ipu-3dgs/