Deep Tech Modular Vision‑Planning System Matches Fine‑Tuned Models Without Robot Data TiPToP demonstrates that a modular, open‑vocabulary planning system can match or surpass a fine‑tuned vision‑language‑action model without any robot‑specific... arXiv Mar 10, 2026 · 3 min read
Deep Tech Time‑dependent drives beat polaron limits in qubit reset. By tailoring a transmon’s Hamiltonian in real time, researchers have shown that polaron‑induced limits on qubit reset can be sidestepped, enabling fidelities... arXiv Mar 10, 2026 · 2 min read
Deep Tech LLM Guides Shift from Tool to Companion for Blind VR Users When alone, blind and low‑vision users treat a large‑language‑model‑powered guide as a tool, but in social settings they anthropomorphise it and engage it as... arXiv Mar 10, 2026 · 3 min read
Deep Tech A complete classification of 2d symmetry protected states with symmetric entanglers The authors demonstrate that every two‑dimensional symmetry‑protected topological state accessible via a symmetric entangler is uniquely labelled by an eleme... arXiv Mar 10, 2026 · 3 min read
Deep Tech Constructive Interference in Neural Networks Explained by Feature Correlations When features in a neural network co‑activate, interference can become constructive, reshaping the geometry of learned representations. arXiv Mar 10, 2026 · 2 min read
Deep Tech Physics‑guided deep learning sharpens global carbon flux estimates TAM‑RL cuts the root‑mean‑square error of terrestrial carbon flux estimates by nearly ten percent across diverse biomes. arXiv Mar 10, 2026 · 3 min read
Deep Tech Approximate QCAs in one dimension using approximate algebras Every approximate quantum cellular automaton on a finite one‑dimensional circle can be replaced by an exact automaton with the same index. arXiv Mar 9, 2026 · 3 min read
Deep Tech Training LLM agents to critique their own actions improves performance and generalization ACT trains large‑language‑model agents to judge which of two actions is better, and this simple change yields measurable gains across multiple benchmarks. arXiv Mar 9, 2026 · 3 min read
Deep Tech Impermanent: A Live Benchmark for Temporal Generalization in Time Series Forecasting Impermanent demonstrates that static train‑test splits in time‑series forecasting routinely inflate reported accuracy, masking the true temporal fragility of... arXiv Mar 9, 2026 · 3 min read
Deep Tech Diffusion Models Can Skip Full‑Resolution Processing The most noisy states of diffusion models contain no more information than a modestly downsampled image, yet are still processed at full resolution. arXiv Mar 9, 2026 · 3 min read
Deep Tech Coupled‑Layer Design Unveils Physical Blueprint for Product Quantum Codes Zhang, Wei, and Tantivasadakarn demonstrate that a coupled‑layer construction can physically assemble any product code from its constituent codes. arXiv Mar 9, 2026 · 3 min read