Publications

Dynamic Neural Graph Encoding of Inference Processes in Deep Weight Space

Di Wu

Huan Liu

Zhixiang Chi

Yuanhao Yu

Konstantinos N. Plataniotis

Yang Wang

The rapid advancements in using neural networks as implicit data representations have attracted significant interest in developing machine l… (see more)earning methods that analyze and process the weight spaces of other neural networks. However, efficiently handling these high-dimensional weight spaces remains challenging. Existing methods often overlook the sequential nature of layer-by-layer processing in neural network inference. In this work, we propose a novel approach using dynamic graphs to represent neural network parameters, capturing the temporal dynamics of inference. Our Dynamic Neural Graph Encoder (DNG-Encoder) processes these graphs, preserving the sequential nature of neural processing. Additionally, we also leverage DNG-Encoder to develop INR2JLS (Implicit Neural Representation to Joint Latent Space) for facilitate downstream applications, such as classifying Implicit Neural Representations (INRs). Our approach demonstrates significant improvements across multiple tasks, surpassing the state-of-the-art INR classification accuracy by approximately 10\% on the CIFAR-100-INR. Our code is available at https://github.com/dddiowww/DNG.

2026-06-09

Transactions on Machine Learning Research (accepted)

Establishment of a tissue culture system with adventitious bud regeneration for the new raspberry germplasm 'autumn–winter yellow raspberry'

Zihan Zhu

Hanqing Zhao

Jinyu Liu

Ye Guo

Chenxing Zhang

Yingyue Li

2026-06-09

Plant (published)

Hierarchical Integration of Predictive Representations of State from General Value Functions

Sonny Jones

Doina Precup

Patrick M. Pilarski

Ashley N Dalrymple

In this work, we investigate how predictive representations of state in the form of continually learned General Value Functions (GVFs) inter… (see more)act with downstream policy networks. Intelligent agents deployed in real-world environments need to adapt to changing conditions in their environment. Adapting to one’s environment requires a model or representation of the environment on which to base decision-making. Models that take the form of predictions and GVFs have been shown to provide temporally abstracted predictive representations of state that can forecast useful elements of an agent's or environment's future behaviour. While GVFs have been concretely deployed in rehabilitation and robotic domains, existing approaches treat predictions as input features into model frameworks, without examining or comparing how best to integrate them into downstream learning processes. In this work, we compare multiple strategies for integrating observations and GVF predictions into another learning architecture: 1) actual observations solely in the input layer, 2) predictions solely in the input layer, 3) actual observations and predictions in the input layer, and 4) actual observations in the input layer and predictions in the later latent representations. We evaluate these strategies in a rehabilitation setting, using GVFs to learn predictive representations of kinetic and kinematic signals collected from wearable sensors on the lower limb during ambulation across varied terrains, and policy networks to classify walking terrain.

2026-06-09

rl-conference.cc/RLC/2026/Workshop/RL_in_Big_Worlds (poster)

Optimizing Dec-POMDP Agent-State Policies via Risk-Seeking Utility

Amit Sinha

Matthieu Geist

Aditya Mahajan

Solving decentralized decision-making problems modeled as Dec-POMDPs is notoriously NEXP-complete, as optimal solutions require policies con… (see more)ditioned on an agent's entire action-observation history. To maintain tractability, it is common to restrict agents to finite-memory models, known as agent-state policies. Although this constrained policy class may not contain the globally optimal solution, finding the highest-performing agent-state policy remains a critical objective for practical applications. Addressing the challenge of planning under bounded memory, we introduce an iterated best-response algorithm that converges monotonically to a local optimum in polynomial runtime relative to the Dec-POMDP model size. To discover superior policies within this restricted memory space, we employ a novel objective that pairs a risk-seeking incentive with conservative policy updates. Our experiments on standard Dec-POMDP benchmarks demonstrate that this approach is competitive with state-of-the-art methods, delivering near-optimal results despite the limited memory.

2026-06-09

rl-conference.cc/RLC/2026/Workshop/RL_in_Big_Worlds (poster)

Representing Time Series as Structured Programs for LLM Reasoning

Jaeho Kim

Changhun Oh

Seokhyun Lee

Irina Rish

Changhee Lee

Large language models (LLMs) have demonstrated strong reasoning and instruction-following capabilities, making them potentially powerful too… (see more)ls for time-series analysis. However, time series lie outside their native textual modality, raising a fundamental question: how should time series be represented so that LLMs can reason about them effectively? Existing work typically serializes raw numerical sequences or fine-tunes pre-trained LLMs on time-series data. These approaches place the burden of extracting temporal structure directly on the LLM, creating a modality mismatch that often degrades performance on long sequences and introduces substantial computational overhead. In this work, we introduce Time-Series-to-Structured-Program representation (T2SP), a deterministic, training-free method that represents a time series as a structured symbolic program. T2SP decomposes time series into trends, periods, and salient events, expressing them in a program-friendly format aligned with the textual and code-like modalities on which LLMs are natively trained. By shifting temporal-structure extraction from the model to the representation itself, T2SP enables off-the-shelf LLMs to leverage their existing reasoning capabilities for time-series understanding. We evaluate T2SP on three reasoning tasks -- editing, captioning, and question answering -- where it consistently improves performance, reduces reasoning time, and lowers failure rates compared with raw-string representations. Our results demonstrate that T2SP provides an effective interface between time series and LLMs.

2026-06-09

arXiv (preprint)

SLowRL: Safe Low-Rank Adaptation for Bridging the Sim-to-Real Gap in Legged Locomotion

Elham Daneshmand

Shafeef Omar

Glen Berseth

Majid Khadiv

Hsiu-Chin Lin

A simulator is, at best, a coarse low-fidelity model of the real world the agent eventually has to act in. Closing this residual gap on hard… (see more)ware is a canonical instance of operating in a big world: the real environment exposes contact dynamics, latencies, and disturbances that the agent was never given the capacity (parameters or data) to model during pretraining. Naive on-hardware fine-tuning is risky --- the policy can damage the robot before it improves --- and full-parameter updates require prohibitive interaction time. We propose SLowRL, a continual fine-tuning framework that confronts this big-world adaptation problem with two complementary forms of capacity limitation: (i) a rank-1 LoRA adapter applied per layer to both actor and critic, restricting each layer's update to a single direction in its image space (

2026-06-09

rl-conference.cc/RLC/2026/Workshop/RL_in_Big_Worlds (poster)

The blueprint of human functional architecture shifts from cognition to anatomy during perturbations of consciousness

Andrea I. Luppi

Dragana Manasova

Justine Y. Hansen

Zhen-Qi Liu

Asa Farahani

Yonatan Sanz Perl

Jakub Vohryzek

Daniel Golkowski

Andreas Ranft

R. Ilg

Denis Jordan

Vincent Bonhomme

Audrey Vanhaudenhuyse

Athéna Demertzi

Océane Jaquet

Mohamed Ali Bahri

Naji Alnagger

Paolo Cardone

Lorina Naci

Adrian M. Owen … (see 9 more)

John Pickard

Guy Williams

Judith Allanson

Enrico Amico

Danilo Bzdok

Jacobo Sitt

David Menon

Emmanuel A. Stamatakis

Bratislav Misic

Consciousness and cognition arise from the ongoing interactions between brain regions. Synchronous fluctuations of fMRI signals may indicate… (see more) that two brain regions perform similar cognitive functions, but neural interactions are also constrained by anatomical connectivity and regions' molecular, cytoarchitectonic, and metabolic profiles. Here we disentangle the respective contributions of ongoing cognition and multimodal neurobiological constraints in shaping functional connectivity. We jointly contextualise haemodynamic FC against eight distinct multimodal representations of the human connectome: (i) structural connectivity from diffusion tractography; (ii) spatial embedding; (iii) similarity of transcriptional profiles from gene expression; (iv) similarity of receptor profiles from Positron Emission Tomography; (v) laminar profile similarity from histology; (vi) correlated electrophysiological activity from magnetoencephalography; (vii) correlated metabolic activity from PET glucose uptake; (viii) coordinated activation across 123 cognitive operations from the NeuroSynth meta-analytic engine. We demonstrate that cognitive co-activation is the dominant predictor of inter-regional fMRI synchrony in the awake human brain, even when quantified using intracranial electrical stimulation. Crucially, this predominance of cognitive co-activation for shaping functional connectivity is systematically obliterated across five datasets of pharmacological and pathological perturbations of consciousness (chronic disorders of consciousness; anaesthesia with sevoflurane, propofol, or ketamine) when cognition is disconnected from the environment or altogether abolished. Altogether, we show that multimodal predictors of functional architecture shift away from cognitive co-activation and toward anatomical-molecular constraints during pharmacological and pathological perturbations of consciousness.

2026-06-09

bioRxiv (accepted)

CIAware-Bench: Benchmarking Control Intervention Awareness Across Frontier LLMs

Joachim Schaeffer

Thomas Jiralerspong

Alexander Panfilov

Guillaume Lajoie

Jonas Geiping

Yoshua Bengio

Roland S. Zimmermann

AI control protocols oversee untrusted models by monitoring their actions and modifying potentially unsafe steps, often using a trusted mode… (see more)l. This partially tampers with the untrusted model's trajectory. If the trusted model detects such an intervention, it may infer properties of the monitor and adapt to evade control. We introduce \textbf{CIAware-Bench}, a benchmark for measuring \textbf{c}ontrol \textbf{i}ntervention (CI) awareness across frontier models. CIAware-Bench tests whether models can distinguish their own trajectories from those modified by a control intervention. The benchmark is comprised of a suite of four task domains (essay writing, BigCodeBench, Bash Arena, and SHADE-Arena), while varying trajectory watermarking, side-task presence, and the control protocol. Evaluating eleven frontier models, we find low to moderate CI awareness under default settings (up to 0.87; random chance balanced binary classification accuracy is 0.5) with substantial variation across task domains and model pairs. Detection is generally easier across model families, suggesting that models exploit provider-specific differences in style or post-training. Overall, CI awareness is not a fixed model-level property, and should be measured for each new model release and deployment scenario. We release CIAware-Bench to track CI awareness and inform control protocols whose interventions are harder to detect.

2026-06-08

arXiv (preprint)

Critical dynamics in spontaneous EEG predict perturbational complexity in disorders of consciousness with measurable evoked responses

Derek Newman

Charlotte Maschke

Jordan O‘Byrne

Michele Colombo

Angela Comanducci

Silvia Casarotto

Giuseppe Citerio

M Rosanova

Marcello Massimini

Karim Jerbi

Stefanie Blain‐Moraes

Abstract Identifying which severely brain-injured patients retain the capacity for consciousness remains a major challenge in neurocritical … (see more)care. The perturbational complexity index (PCI) provides a reliable assessment of consciousness capacity, but its reliance on transcranial magnetic stimulation and EEG (TMS-EEG) limits bedside scalability. PCI and brain criticality capture complementary dimensions of brain dynamics: PCI quantifies the complexity of the brain’s evoked response to perturbation, whereas criticality characterizes the intrinsic organization of spontaneous activity. Here, we tested whether resting-state EEG signatures of criticality predict PCI max in disorders of consciousness, extending prior findings from anesthesia to severe brain injury. In 26 patients with vascular, traumatic, or anoxic brain injury, multivariate criticality related features did not generalize PCI max prediction across the full heterogeneous cohort. However, criticality features predicted PCI max when analyses were restricted to non-anoxic patients and when restricting analyses to patients with non-zero PCI max values. These findings suggest that spontaneous criticality measures index the brain’s intrinsic dynamical regime that supports complex perturbational responses, while their correspondence with PCI max depends on whether the injured brain retains sufficient capacity to sustain large-scale evoked responses. Together, our results extend the relationship between resting-state criticality and evoked perturbational complexity to disorders of consciousness and support the development of stratified EEG measures in severe brain injury.

2026-06-08

bioRxiv (accepted)

Human learning of noninvasive brain–computer interfaces via manifold geometry

Erica L. Busch

E. Chandra Fincke

Guillaume Lajoie

Smita Krishnaswamy

Nicholas B. Turk‐Browne

2026-06-08

Nature Neuroscience (published)

Rank Collapse, Fixed Points, and the Renormalization Group Structure of MLP Residual Networks

Parviz Haggi-Mani

Irina Rish

The analogy between deep neural network forward passes and renormalization group (RG) flows has been repeatedly noted in the literature, but… (see more) existing treatments remain qualitative: depth is described as a coarse-graining scale, attention is likened to a partition function, and representations are said to flow toward fixed points. No existing work has defined a measurable RG order parameter, tested it under controlled variation of the input distribution, or made quantitative predictions that are empirically verified. We study the simplest architecture for which the analogy is tractable: a pure MLP residual stack trained on masked token prediction over synthetic Markov chain sequences with known spectral properties. We report three findings. (i) The effective rank of the residual stream decreases monotonically with depth after training, consistent with progressive integration of irrelevant degrees of freedom. (ii) This rank collapse is selective: it occurs for chains with short correlation length approximately 1 but is absent for chains with long correlation length approximately 7, measured at the position level to control for mean-pooling artifacts. The network preserves exactly the degrees of freedom relevant to the prediction task, the content of the RG relevance criterion. (iii) Inter-layer kernel drift is concentrated at one or two specific transitions, with the remainder of the network near a fixed point, consistent with a discrete fixed-point plateau. Together these findings constitute the first quantitative, position-level evidence that MLP residual networks implement a selective coarse-graining procedure governed by the spectral structure of the input distribution.

2026-06-08

arXiv (preprint)