7 papers
Attention Manifolds: Steering or Blocking Language Models by Editing Learned B-Spline Surfaces
Naveen Mysore
In standard transformer attention, a source token sends the same value vector to every receiver. The query determines \emph{how much} to attend but not \emph{what} to extract. This…
Recovering Governing Dynamics from Distributed Observations via Exact Spline Merging
Naveen Mysore
Scientific observations are frequently distributed across locations, time periods, and institutions. Combining such observations into a continuous, differentiable field enables rec…
InKAN: B-Spline KANs via Truncated Power Form
Naveen Mysore
Kolmogorov-Arnold Networks (KANs) place learnable B-spline activations on network edges rather than fixed activations on nodes. The standard Cox-de Boor recursion evaluates these a…
Temporal Functional Circuits: From Spline Plots to Faithful Explanations in KAN Forecasting
Naveen Mysore
Unlike MLPs, Kolmogorov-Arnold Networks (KANs) expose explicit learnable edge functions on every connection, enabling mechanistic explanation in time-series forecasting. This paper…
DecompKAN: Decomposed Patch-KAN for Long-Term Time Series Forecasting
Naveen Mysore
Accurate time series forecasting in scientific domains such as climate modeling, physiological monitoring, and energy systems benefits from both competitive predictions and model t…
Prediction-Based Markov Violation Scores for Detecting Non-Markovian Observations in Reinforcement Learning
Naveen Mysore
Reinforcement learning algorithms assume that observations satisfy the Markov property, yet real-world sensors frequently violate this assumption through correlated noise, latency,…
Quantifying First-Order Markov Violations in Noisy Reinforcement Learning: A Causal Discovery Approach
Naveen Mysore
Reinforcement learning (RL) methods frequently assume that each new observation completely reflects the environment's state, thereby guaranteeing Markovian (one-step) transitions.…