A Bayesian Approach for Task-Specific Next-Best-View Selection with Uncertain Geometry
Bayesian framework for active view selection in 3D reconstruction using posterior inference over implicit surfaces.
Search the full wire by company, model, lab, or keyword. Every story we have ever aggregated.
Bayesian framework for active view selection in 3D reconstruction using posterior inference over implicit surfaces.
Psychometric analysis of 50 LLMs identifies phenomenal experience as primary variance axis via Pinocchio dimension.
Fully convolutional neural network for chemical-mechanical polishing modeling in IC manufacturing using white light interferometry.
Systematic review of jailbreak attack and defense methods for LLMs with critique of narrow evaluation metrics like attack success rate.
Graph-SND: sparse-graph generalization of System Neural Diversity metric for multi-agent RL, reducing quadratic-time computation to O(|E|) with unbiased estimation.
Activation-based steering provides control of LLM behavior at inference time, but the dominant paradigm reduces each concept to a single direction whose geometry is left largely unexamined. Rather than selecting a single steering direction, we use conceptors: soft projection…
Geometric continuity in deep networks explained by residual connections and symmetry-breaking nonlinearities coordinating weight updates across layers.
ReshapeOT improves optimal transport for distribution shifts by reshaping ground metrics using observed sample displacements.
Theoretical analysis of batch normalization's effect on geometry of piecewise-affine networks during training via hyperplane switching.
Evolving Idea Graphs (EIG), a multi-agent LLM framework using learnable graph edits for scientific ideation with novelty, feasibility, clarity metrics.
Benchmark comparison shows Gemma 4 31B trades inference speed for token efficiency vs Qwen 3.6/5 27B; Qwen optimizes for metrics, Gemma for throughput.
Task vector arithmetic on BEATs encoders composes 661-species bioacoustic classifier without data sharing; task vectors near-orthogonal, geometry aligns with acoustic niche hypothesis.
TRACE: engineering framework for trustworthy agentic AI in critical domains combining reference architecture, trust metrics, and bounded human supervision.
Theoretical characterization of Bayes-consistency for learning with general metric losses in the realizable setting.
IConFace uses identity-structure asymmetric conditioning for reference-aware face restoration under severe degradation.
Perceptual Flow Network mitigates vision-language model hallucination by eschewing rigid geometric priors for interpretable perceptual trajectories.
OphMAE is a multimodal foundation model bridging volumetric and planar ophthalmic imaging for adaptive diagnosis in resource-limited settings.
Fréchet-based random-effects algorithm for non-Euclidean data in metric spaces; extends mixed-effects models beyond Hilbert spaces.
Spectral Model eXplainer: domain-grounded explainability framework for chemometrics ML models operating on spectral data.
Stratechery analysis: Google's stock outperformed Meta's despite weaker core metrics; Google's AI monetization strategy (including Anthropic investment) cited as key driver.
GeoSAE framework uses geometric priors and sparse autoencoders to interpret clinical information in brain MRI foundation models while avoiding feature collapse.
Hybrid architecture combining low-bitrate HEVC video and JPEG ROI stills for bandwidth-constrained robotic vision perception.
TMD-Bench introduces evaluation metrics for text-driven music-dance co-generation, addressing rhythm-choreography coupling beyond generic audiovisual consistency.
Calibrated Size Ratio (CSR) replaces Expected Calibration Error for confidence calibration, addressing ECE's inability to detect overconfidence risk.
Geometric unlearning method removes specific content from LLMs without full training corpus access, balancing privacy and model utility.
GEASS steering mitigates object hallucination in vision-language models by asymmetric caption weighting without retraining.
Probe-Geometry Alignment surgically removes memorization traces from unlearned LLMs via cross-sequence detection without capability loss.
I suck at 3D modeling, so I was excited to test the Blender Connector to see if Claude could help reproduce basic geometry that I struggle with. I asked it to reproduce a sci-fi space shuttle design from a piece of artwork. I'm happy to report that *one* of us was pleased with…
**New arxiv paper just landed that's worth reading if you're interested in stylometry, AI revision, or the prose-writing strand of the 4.7 discussion.** Berkeley researcher Tom van Nuenen ran 300 personal narratives through three frontier models (Claude-class, ChatGPT-class…
SAVGO RL algorithm embeds state-action pairs with cosine similarity to shape policy updates, improving sample efficiency in continuous control.