Chem-GMNet uses sphere-native embeddings, DualSKA attention, and SH-FFN layers to match or beat ChemBERTa-2 on MoleculeNet tasks with fewer parameters and sometimes no pretraining.
hub Canonical reference
nGPT: Normalized transformer with representation learning on the hypersphere
Canonical reference. 100% of citing Pith papers cite this work as background.
10
Pith papers citing it
Background
100% of classified citations
hub tools
citation-role summary
background 6
citation-polarity summary
roles
background 6polarities
background 6representative citing papers
Attention sinks arise from variance discrepancy in self-attention value aggregation, amplified by super neurons and first-token dimension disparity, and can be mitigated by head-wise RMSNorm to accelerate pre-training convergence.
Manifold constraints via the new MACRO optimizer independently bound activation scales and enforce rotational equilibrium in LLM pre-training, subsuming RMS normalization and decoupled weight decay while delivering competitive performance with convergence guarantees.
Polaris learns hierarchical concepts via coupled orbital polar embeddings on hyperspheres that separate meaning from structure using tangent projections, exponential maps, and asymmetric objectives, yielding up to 19-point gains in top-K retrieval.
FlashSAC improves training speed and final performance of off-policy RL on high-dimensional robot tasks by reducing update frequency, increasing model scale, and bounding norms to limit critic error accumulation.
Strong superposition causes neural loss to scale as the inverse of model dimension due to geometric feature overlaps, explaining scaling laws for broad frequency distributions.
Normalized Matching Transformer enforces unit-norm embeddings at every Transformer layer and trains with InfoNCE plus hyperspherical uniformity loss, reaching new state-of-the-art accuracy on PascalVOC and SPair-71k while converging faster than prior matching networks.
Proximal stochastic spectral preconditioning converges for nonconvex constrained objectives under heavy-tailed noise, with a variance-reduced version achieving faster rates and a refined analysis of Muon iterations.
DyT improves validation loss 27% at 64M params/1M tokens but worsens it 19% at 118M tokens, with saturation levels predicting the sign of the effect.
citing papers explorer
-
Chem-GMNet: A Sphere-Native Geometric Transformer for Molecular Property Prediction
Chem-GMNet uses sphere-native embeddings, DualSKA attention, and SH-FFN layers to match or beat ChemBERTa-2 on MoleculeNet tasks with fewer parameters and sometimes no pretraining.
-
The Structural Origin of Attention Sink: Variance Discrepancy, Super Neurons, and Dimension Disparity
Attention sinks arise from variance discrepancy in self-attention value aggregation, amplified by super neurons and first-token dimension disparity, and can be mitigated by head-wise RMSNorm to accelerate pre-training convergence.
-
Demystifying Manifold Constraints in LLM Pre-training
Manifold constraints via the new MACRO optimizer independently bound activation scales and enforce rotational equilibrium in LLM pre-training, subsuming RMS normalization and decoupled weight decay while delivering competitive performance with convergence guarantees.
-
Polaris: Coupled Orbital Polar Embeddings for Hierarchical Concept Learning
Polaris learns hierarchical concepts via coupled orbital polar embeddings on hyperspheres that separate meaning from structure using tangent projections, exponential maps, and asymmetric objectives, yielding up to 19-point gains in top-K retrieval.
-
FlashSAC: Fast and Stable Off-Policy Reinforcement Learning for High-Dimensional Robot Control
FlashSAC improves training speed and final performance of off-policy RL on high-dimensional robot tasks by reducing update frequency, increasing model scale, and bounding norms to limit critic error accumulation.
-
Superposition Yields Robust Neural Scaling
Strong superposition causes neural loss to scale as the inverse of model dimension due to geometric feature overlaps, explaining scaling laws for broad frequency distributions.
-
Normalized Matching Transformer
Normalized Matching Transformer enforces unit-norm embeddings at every Transformer layer and trains with InfoNCE plus hyperspherical uniformity loss, reaching new state-of-the-art accuracy on PascalVOC and SPair-71k while converging faster than prior matching networks.
-
Constrained Stochastic Spectral Preconditioning Converges for Nonconvex Objectives
Proximal stochastic spectral preconditioning converges for nonconvex constrained objectives under heavy-tailed noise, with a variance-reduced version achieving faster rates and a refined analysis of Muon iterations.
-
When Does Removing LayerNorm Help? Activation Bounding as a Regime-Dependent Implicit Regularizer
DyT improves validation loss 27% at 64M params/1M tokens but worsens it 19% at 118M tokens, with saturation levels predicting the sign of the effect.
- Robust Filter Attention: Self-Attention as Precision-Weighted State Estimation