REVIEW 24 cited by
Representation Degeneration Problem in Training Natural Language Generation Models
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Representation Degeneration Problem in Training Natural Language Generation Models
read the original abstract
We study an interesting problem in training neural network-based models for natural language generation tasks, which we call the \emph{representation degeneration problem}. We observe that when training a model for natural language generation tasks through likelihood maximization with the weight tying trick, especially with big training datasets, most of the learnt word embeddings tend to degenerate and be distributed into a narrow cone, which largely limits the representation power of word embeddings. We analyze the conditions and causes of this problem and propose a novel regularization method to address it. Experiments on language modeling and machine translation show that our method can largely mitigate the representation degeneration problem and achieve better performance than baseline algorithms.
Forward citations
Cited by 24 Pith papers
-
STRABLE: Benchmarking Tabular Machine Learning with Strings
A new corpus of 108 mixed string-numeric tables shows that advanced tabular learners with basic string embeddings perform well on most real-world data, while large LLM encoders help on free-text heavy tables.
-
Channel Location Constrains the Auditability of Subliminal Learning
Auditability of subliminal learning is constrained by channel location, with initialization-dependent body channels allowing pre-training screens while vocabulary geometry and conditional body channels evade them.
-
How Does Attention Help? Insights from Random Matrices on Signal Recovery from Sequence Models
Attention pooling produces a free-multiplicative-convolution bulk spectrum and two phase transitions for signal recovery; optimal weights are the top eigenvector of the positional correlation matrix R.
-
LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale
LLM.int8() performs 8-bit inference for transformers up to 175B parameters with no accuracy loss by combining vector-wise quantization for most features with 16-bit mixed-precision handling of systematic outlier dimensions.
-
SpecFormer: Mitigating Embedding and Attention Collapse via Spectral-Aware Transformer for Recommendation
A spectral-softening Transformer that flattens embedding singular values reduces attention collapse and improves CTR prediction accuracy in recommender systems.
-
Reliability Scaling Laws for Quantized Large Language Models
Reliability of quantized LLMs peaks nonlinearly at 4-bit precision under fixed total model bits, while accuracy scales monotonically, and quantization can improve robustness to natural perturbations.
-
Decoupled Residual Quantization for Robust Semantic IDs in Recommendation
Presents a diagnostic framework for semantic ID tokenizer failures using overlap and capacity metrics and proposes DRQ to separate geometry from distribution matching.
-
NITP: Next Implicit Token Prediction for LLM Pre-training
NITP augments next-token prediction with cosine alignment to stop-gradient shallow-layer features of the next token, improving geometry and downstream scores at ~2% extra training FLOPs.
-
NITP: Next Implicit Token Prediction for LLM Pre-training
NITP augments standard next-token prediction with implicit semantic prediction in representation space using shallow-layer self-supervision, reporting consistent downstream gains on 0.5B-9B models including 5.7% on MM...
-
Layer-wise Representation Dynamics: An Empirical Investigation Across Embedders and Base LLMs
LRD framework with Frenet, NRS, and GFMI metrics shows layer-wise structure in 31 models provides usable signal for model selection and pruning on MTEB tasks.
-
Geometric Decoupling: Diagnosing the Structural Instability of Latent
Latent diffusion models exhibit geometric decoupling where curvature in out-of-distribution generation is misallocated to unstable semantic boundaries instead of image details, identifying geometric hotspots as the st...
-
Towards Faster Language Model Inference Using Mixture-of-Experts Flow Matching
Mixture-of-experts flow matching enables non-autoregressive language models to achieve autoregressive-level quality in three sampling steps, delivering up to 1000x faster inference than diffusion models.
-
Geometry-Aware Localized Watermarking for Copyright Protection in Embedding-as-a-Service
GeoMark decouples local watermark triggering from centralized ownership attribution using geometry-separated anchors and adaptive neighborhoods to improve robustness against paraphrasing, dimension changes, and cluste...
-
Revisiting Anisotropy in Language Transformers: The Geometry of Learning Dynamics
Anisotropy in language transformers arises because training amplifies tangent directions, with activation-based low-rank proxies capturing unusually large gradient energy and anisotropy share compared to controls.
-
Maximizing Local Entropy Where It Matters: Prefix-Aware Localized LLM Unlearning
PALU improves LLM unlearning by restricting entropy maximization to sensitive prefixes and top-k logits, achieving better forgetting with less utility loss.
-
Maximizing Local Entropy Where It Matters: Prefix-Aware Localized LLM Unlearning
PALU shows that unlearning only needs local intervention—the first few tokens of the sensitive span and the top-k logits—not full-sequence, full-vocabulary suppression.
-
Scaling Point-in-Time Language Models
Scaling point-in-time decoder-only LMs to 4B parameters on 1T chronologically filtered FineWeb tokens largely closes the quality gap to unrestricted open models while preserving temporal validity.
-
How to deal with machine learning bias in economic history
The paper guides ML use in economic history, identifies systematic prediction bias that distorts coefficients, and shows debiasing via small expert-labeled samples can correct it while preserving scale.
-
NITP: Next Implicit Token Prediction for LLM Pre-training
NITP adds dense supervision from shallow model layers to predict implicit next-token semantics, yielding consistent downstream gains on 0.5B-9B models with ~2% extra training FLOPs.
-
HyperLens: Quantifying Cognitive Effort in LLMs with Fine-grained Confidence Trajectory
HyperLens reveals that deeper transformer layers magnify small confidence changes into fine-grained trajectories, allowing quantification of cognitive effort where complex tasks demand more and standard SFT can reduce it.
-
Scaling Point-in-Time Language Models
Scaling point-in-time LLMs to 4B parameters and 1T temporally filtered tokens narrows the gap to unrestricted models to about 8–11 average points and yields positive out-of-sample Sharpe ratios from news embeddings.
-
Smarter, Better, Faster, Longer: A Modern Bidirectional Encoder for Fast, Memory Efficient, and Long Context Finetuning and Inference
ModernBERT is a new bidirectional encoder model achieving SOTA performance on diverse classification and retrieval benchmarks while offering superior speed and memory efficiency for long-context inference.
-
Learning to Prompt: Improving Student Engagement with Adaptive LLM-based High-School Tutoring
An adaptive prompt router trained in simulation and deployed with high-school students improves exercise conversion to 28.1% and cuts conversation length by about 3 turns compared with static baselines.
-
Elucidating Representation Degradation Problem in Diffusion Model Training
Diffusion models suffer representation degradation at high noise due to recoverability mismatch; ERD mitigates this by dynamic optimization reallocation, accelerating convergence across backbones.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.