Pretraining and alignment induce asymmetric geometric traces in transformer weights because alignment updates concentrate in read pathways due to activation covariance while write pathways inherit less structure from alignment losses.
Breaking the softmax bottleneck: A high-rank rnn language model.arXiv preprint arXiv:1711.03953
9 Pith papers cite this work. Polarity classification is still indexing.
9
Pith papers citing it
abstract
We formulate language modeling as a matrix factorization problem, and show that the expressiveness of Softmax-based models (including the majority of neural language models) is limited by a Softmax bottleneck. Given that natural language is highly context-dependent, this further implies that in practice Softmax with distributed word embeddings does not have enough capacity to model natural language. We propose a simple and effective method to address this issue, and improve the state-of-the-art perplexities on Penn Treebank and WikiText-2 to 47.69 and 40.68 respectively. The proposed method also excels on the large-scale 1B Word dataset, outperforming the baseline by over 5.6 points in perplexity.
representative citing papers
Probabilistic circuits have an output bottleneck with convex probability combinations and a context bottleneck limited to fixed vtree-aligned partitions, making them less expressive than transformers for language data with heterogeneous dependencies, though decomposable PCs are strictly more capable
XLNet is a generalized autoregressive pretraining method that learns bidirectional contexts via permutation-based factorization and outperforms BERT on 20 NLP tasks.
Subliminal steering transfers complex behavioral biases and the underlying steering vector through fine-tuning on innocuous data, achieving higher precision than prior prompt-based methods.
SVD on the lm_head weight matrix of transformers reveals interpretable vocabulary clusters that indicate training data composition, model differences, and ethical concerns in models like GPT-OSS, Gemma, and Qwen.
MoBA routes attention over blocks via MoE-style gating to enable dynamic, bias-light long-context attention that matches full attention performance at lower cost.
Compressive Transformer sets new records on WikiText-103 (17.1 ppl) and Enwik8 (0.97 bpc) via memory compression and introduces the PG-19 long-range language benchmark.
BitLM replaces per-token softmax with bitwise continuous diffusion inside causal blocks to generate multiple tokens in parallel while preserving autoregressive structure.
DiT-ST converts complete-text captions into split-text primitives via LLMs and injects them hierarchically across denoising stages to reduce semantic confusion in DiT-based text-to-image generation.
citing papers explorer
-
Where Pretraining writes and Alignment reads: the asymmetry of Transformer weight space
Pretraining and alignment induce asymmetric geometric traces in transformer weights because alignment updates concentrate in read pathways due to activation covariance while write pathways inherit less structure from alignment losses.
-
The Expressivity Boundary of Probabilistic Circuits: A Comparison with Large Language Models
Probabilistic circuits have an output bottleneck with convex probability combinations and a context bottleneck limited to fixed vtree-aligned partitions, making them less expressive than transformers for language data with heterogeneous dependencies, though decomposable PCs are strictly more capable
-
XLNet: Generalized Autoregressive Pretraining for Language Understanding
XLNet is a generalized autoregressive pretraining method that learns bidirectional contexts via permutation-based factorization and outperforms BERT on 20 NLP tasks.
-
Subliminal Steering: Stronger Encoding of Hidden Signals
Subliminal steering transfers complex behavioral biases and the underlying steering vector through fine-tuning on innocuous data, achieving higher precision than prior prompt-based methods.
-
Check Your LLM's Secret Dictionary! Five Lines of Code Reveal What Your LLM Learned (Including What It Shouldn't Have)
SVD on the lm_head weight matrix of transformers reveals interpretable vocabulary clusters that indicate training data composition, model differences, and ethical concerns in models like GPT-OSS, Gemma, and Qwen.
-
MoBA: Mixture of Block Attention for Long-Context LLMs
MoBA routes attention over blocks via MoE-style gating to enable dynamic, bias-light long-context attention that matches full attention performance at lower cost.
-
Compressive Transformers for Long-Range Sequence Modelling
Compressive Transformer sets new records on WikiText-103 (17.1 ppl) and Enwik8 (0.97 bpc) via memory compression and introduces the PG-19 long-range language benchmark.
-
BitLM: Unlocking Multi-Token Language Generation with Bitwise Continuous Diffusion
BitLM replaces per-token softmax with bitwise continuous diffusion inside causal blocks to generate multiple tokens in parallel while preserving autoregressive structure.
-
Enhancing Text-to-Image Diffusion Transformer via Split-Text Conditioning
DiT-ST converts complete-text captions into split-text primitives via LLMs and injects them hierarchically across denoising stages to reduce semantic confusion in DiT-based text-to-image generation.