FocalPolicy: Frequency-Optimized Chunking and Locally Anchored Flow Matching for Coherent Visuomotor Policy
FocalPolicy is a foresight-aware visuomotor policy that combines Frequency-Optimized Chunking with Locally Anchored flow matching . It combines time-domain alignment within the proximal actions while regularizing frequency-domain structure over multiple future action chunks to improve cross-chunk coherence .
π’ Applied
Unsupervised Domain Shift Detection with Interpretable Subspace Attribution
π‘ This tool helps researchers detect subtle differences in dataset probability distributions by identifying localised density anomalies in high-dimensional feature spaces.
This tool detects subtle differences in dataset probability distributions by identifying localised density anomalies in high-dimensional feature spaces.
domain shiftfeature spaceanomaly detection
π’ Applied
Ti-iLSTM: A TinyDL Approach for Logic-Level Anomaly Detection in Industrial Water Treatment Systems
π‘ This system can manipulate process behaviour without creating obvious devices outliers, which is a form of logic-layer deception that can preserve numerically plausible measurements while breaking expected cause-and-effect relationships.
This system can manipulate process behaviour without creating obvious devices outliers, which is a form of logic-layer deception that can preserve numerically plausible measurements while breaking expected cause-and-effect relationships.
cybersecurityanomaly detectionwater treatment
π’ Applied
Navigating Potholes with Geometry-Aware Sharpness Minimization
π‘ This method uses a learned preconditioner to address the issue of ignoring loss geometry, making it applicable to practical applications.
LLQR+SAM combines sharpness-aware minimization with a learned preconditioner to address the issue of ignoring loss geometry.
optimizationmachine learninggeometryminimization
π’ Applied
Multi-Fidelity Flow Matching: Cascaded Refinement of PDE Solutions
π‘ MFFM is a refinement technique that uses conditional flow matching to improve parametric solutions by conditioning on data-driven source distributions.
Multi-Fidelity Flow Matching is a refinement framework that uses cascading refinement to improve parametric PDE solutions by conditioning on data-driven source distributions.
flow_matchingpdeconditional_flow_matchingresidual_refinementparametric_solutions
π’ Applied
MIND: Decoupling Model-Induced Label Noise via Latent Manifold Disentanglement
π‘ MIND helps prevent model-induced label noise by decoupling the noise manifold into subspace-dependent components.
MIND decouples model-induced label noise using Latent Manifold Disentanglement to separate noise components.
neural networksdeep learningmodel-induced noisedisentanglementlatent manifold
π’ Applied
Looped SSMs: Depth-Recurrence and Input Reshaping for Time Series Classification
π‘ Looped SSMs are a recurrent model family that can be improved by repeating blocks across layers and reshaping input data.
Looped SSMs are a recurrent model family that can be improved by repeating blocks across layers and reshaping input data.
recurrentssmmodelrepetitioninputreshaping
π’ Applied
Judge Circuits
π‘ This study investigates how different models are graded by analyzing internal mechanisms using position-aware edge attribution patching.
This research explains how different model outputs are graded by examining internal mechanisms using position-aware edge attribution patching.
llmattentionmodeljudginggradingpatching
π’ Applied
Clock-state olfactory search in turbulent flows using Q-learning: The geometry of plume recovery
π‘ This study uses Q-learning to train an agent to recover plumes in turbulent flows, combining insect behaviors to improve recovery.
This research uses Q-learning to train an agent to recover plumes in turbulent flows, combining insect behaviors to improve recovery.
learningcontrolbiologyplumeturbulence
π’ Applied
Shapley Neuron Values for Continual Learning: Which Neurons Matter Most?
π‘ Shapley Neuron Valuation is a method for continual learning that quantifies Neuron importance by freezing important neurons while keeping others plastic, enabling buffer-free continual learning without expanding architecture.
Shapley Neuron Valuation quantifies Neuron importance in continual learning by freezing important neurons while keeping others plastic.
continual learningneuron importancelearning theory
π’ Applied
StippleDiffusion: Capacity-Constrained Stippling using Controlled Diffusion
π‘ This method uses diffusion to create stipple patterns while ensuring the local density matches a learned prior and respects a continuous image-defined capacity constraint.
A diffusion-based sampler for stipple patterns that satisfies a learned local point-distribution prior and a continuous image-defined capacity constraint.
diffusionimage processingstipplepattern generation
π’ Applied
Continual Learning of Domain-Invariant Representations
π‘ This paper describes a method that trains models to learn domain-invariant representations, preventing them from learning spurious domain-specific cues
Continual learning of domain-invariant representations addresses the problem of shortcut learning by training models to learn domain-invariant representations
continual learningdomain invariant representationsshortcut learning
π’ Applied
GOMA: Toward Structure-Driven Multimodal Alignment from a Graph Signal Smoothing Perspective
π‘ GOMA is a structure-driven post-alignment framework that decouples three key design choices to improve multimodal alignment.
GOMA is a structure-driven multimodal alignment framework that decouples three key design choices: where messages flow, how multimodal evidence propagates, and smoothing depth.
multimodalgraphcommunicationalignment
π’ Applied
Scalable neuromorphic computing from autonomous spiking dynamics in a clockless reconfigurable chip
π‘ This system uses autonomous spiking dynamics in clockless digital circuits to create networks of interacting neurons with configurable weights, enabling scalable neuromorphic computing.
This system uses autonomous spiking dynamics in clockless digital circuits to create networks of interacting neurons with configurable weights, enabling scalable neuromorphic computing.
neurosciencecomputer architecturedistributed systemsspiking neural networksfield-programmable gate arraysneuromorphic computing
π’ Applied
A numerical study into neural network surrogate model performance for uncertainty propagation
π‘ This research explores using neural networks to approximate complex physical solutions for stochastic problems, potentially lowering the computational cost of repeated simulations.
This study examines how neural network surrogate models can reduce repeated expensive forward model evaluations for solving boundary value problems in stochastic scenarios.
stochastic problemsboundary value problemssurrogate modelingnumerical analysis
π’ Applied
SAFE Quantum Machine Learning with Variational Quantum Classifiers
π‘ This research explores using quantum machine learning to solve optimization problems, specifically a variational quantum classifier that operates on high-dimensional deep representations.
This paper proposes a variational quantum classifier for optimization problems using amplitude encoding and a learnable classical pre-encoding layer, with model reliability assessed via SAFE-AI metrics.
optimizationmachine learningquantum machine learningvariational quantum classifiers
π’ Applied
Explainable AI Isn't Enough! Rethinking Algorithmic Contestability
π‘ Individuals can respond to negative decisions made by opaque machine learning systems, and we propose an operational definition of contestability as a complement to recourse.
Machine learning systems increasingly make life-changing decisions about individuals, and we propose an operational definition of contestability as a complement to recourse.
aimachine learningalgorithmiccontestability
Mind Dreamer: Untethering Imagination via Active Latent Intervention on Latent Manifolds
Model-Based Reinforcement Learning (MBRL) leverages latent imagination for sample efficiency, yet remains constrained by Historical Tethering . We propose Mind Dreamer (MD), a framework that operationalizes Active Latent Intervention (ALI) to transcend Markovian continuity . MD reformulates discovery as the minimization of a global Relay Manifold Expected Free Energy (R-EFE)
π’ Applied
Variational Autoregressive Networks with probability priors
π‘ This paper introduces a method for training neural networks that can incorporate physical priors to improve performance, using a prior probability distribution as a starting point.
This paper proposes a framework for training Variational Autoregressive Networks with probability priors, enhancing performance by incorporating physical priors.
machine learningneural networksphysics
π’ Applied
Adaptive Outer-Loop Control of Quadrotors via Reinforcement Learning
π‘ This research introduces a new control method for quadrotors that uses Reinforcement Learning to adapt to real-world disturbances.
This paper proposes an adaptive control architecture for quadrotors using Reinforcement Learning to handle real-world perturbations.
airoboticscontrolflight control
π’ Applied
Testing properties of trees in graphical models with covariance queries
π‘ Testing properties of graphs in graphical models using covariance queries is a specialized statistical problem that helps researchers analyze complex systems by examining relationships between variables.
Testing properties of graphs in graphical models using covariance queries is a specialized statistical problem that helps researchers analyze complex systems by examining relationships between variables.
machine learningstatisticsgraph theorydata science
π’ Applied
Entropy-Based Characterisation of the Polarised Regime in Latent Variable Models
π‘ This paper proposes a simple method to identify active dimensions in variational models using entropy-based criteria.
This paper proposes a simple information-theoretic classification method based on entropy to identify active dimensions in variational models.
entropyvariancelatent variables
π’ Applied
Imperfect World Models are Exploitable
π‘ We define exploitable models as those where one policy should be preferred over another while the environment's true model suggests the reverse, similar to how reward hacking is characterized.
We propose a novel definition of model exploitation in reinforcement learning, comparing it to reward hacking to characterize exploitable models.
reinforcement_learningmodel_exploitationexploitation
A Retrieval-Enhanced Transformer for Multi-Step Port-of-Call Sequence Prediction in Global Liner Shipping
Connectivity-Constrained and Retrieval-Enhanced (CCRE) deep learning framework proposes a deep-learning framework . CCRE achieves a 72.3% first-destination accuracy and a 61.4% average three-step accuracy, outperforming baselines like CatBoost and LSTM .
π’ Applied
LoCO: Low-rank Compositional Rotation Fine-tuning
π‘ LoCO lets researchers fine-tune large models with low-rank rotations, making it practical for tasks like NLP and CV.
LoCO enables efficient fine-tuning of large models using low-rank rotations.
fine-tuningrotationslarge_models