Spotlight

Papers

Field

Top tools here

Open problems in Graph learning

Quantization error propagation in recurrent states1 paper raises it · 1 addresses it
Low-precision recurrent states in linear attention degrade accuracy due to error accumulation over time and across dimensions.
- STEPQuant: When and Where Errors Matter in Delta-Rule Recurrent State Quantizationraises · addresses
Train-rollout discrepancy in low-precision RL1 paper raises it · 1 addresses it
Quantization in RL training and rollout paths can diverge, reducing the effectiveness of low-precision methods.
- TRACE: Rollout-Guided Quantization-Aware Training for FP4 Reinforcement Learning of MoE Language Modelsraises · addresses
Distribution shift in pruning for decoding1 paper raises it · 1 addresses it
Pruning guided by natural sequences fails to account for self-generated token distributions during decoding, hurting performance.
- SparseDecoding: Decoding-Aware Pruning for Accurate and Efficient LLM Inferenceraises · addresses
Lack of benchmarks for graph decision-making in System One models1 paper raises it · 1 addresses it
Existing benchmarks do not adequately evaluate System One models on graph tasks, limiting progress in this area.
- GraphDecisionBench: Benchmarking System One Models on Graph Tasksraises · addresses
Calibration dependency in diffusion model quantization1 paper raises it · 1 addresses it
Post-training quantization for diffusion models often relies on calibration data, which is impractical due to shifting activation distributions.
- SoloQ: Calibration-Free Quantization for Diffusion Language Modelsraises · addresses

Named by an AI from these papers' abstracts.

Open problems

Recently funded

Topics in this field

graph learning · graph neural · graph neural networks256 papers41 this week
gradient descent · learning rate · neural networks209 papers35 this week
neural networks · neural network · arnold networks143 papers33 this week
memory efficient · optimizers · low rank adaptation151 papers24 this week
bit quantization · quantization aware training · quantization aware147 papers16 this week
linear attention · softmax attention · transformer attention104 papers13 this week

Topics in this field

Active contributors · 90 days

Active contributors

Activity · 12 weeks

Coming with more sources: repos

Activity

★ Spotlight this month

arXiv cs.LGCodeOct 14▲ 5

Your Transformer Is Secretly Elastic

Walter Simoncini, Dheeraj Varghese, Gertjan J. Burghouts, Cees G. M. Snoek

Scaling up transformers has produced models with broad capabilities, deployed across heterogeneous hardware and fluctuating workloads. Yet, model families offer only a few fixed sizes (e.g., S, B, L), forcing users to commit to a model based on hardware and latency constraints. This rigidity is often suboptimal under varying query loads. We introduce a retraining-free method that elastifies pre-trained transformers into a single model that adapts on-the-fly across a continuum of computational budgets. Our key contribution is a differentiable relaxation of the functional unit-ranking problem: unlike prior work, which optimizes a single importance weight per MLP block or attention head, we rank every individual MLP neuron and attention head jointly, on a common scale, by minimizing the model's loss under soft pruning across many sparsity targets, all at once. Across 27 vision and text transformer encoders ranging from 22M to 8B parameters, our method yields smooth degradation up to 60% sparsity and outperforms one-shot baselines under deep pruning. For example, pruning DINOv3 ViT-H+/16 to 50% sparsity removes 427M parameters while reducing ImageNet-1k linear-probing accuracy by…

PDF arXiv page Details & code

Spotlight

Recent papers

All in Research →

arXiv cs.CLCodeOct 19▲ 79
SpecMRL: Redesign Matryoshka Representation Learning within Parametric Spectrum
Zhongxin Chen, Songhao Wu, Heng Cui, Cong Li, Xiao Zhou +1
Matryoshka Representation Learning (MRL) is designed to support flexible representation sizes by jointly optimizing nested prefixes. While effective, this design lacks an explicit constraint to ensure that the leading portion provides a suitable compact encoding. To address this, we propose SpecMRL, which reorganizes the feature space to enable more effective prefix readout. Specifically, we extend MRL with a parameter-derived spectral basis organized with feature importance. SpecMRL otherwise follows standard MRL, with prefix readout performed in the learned spectral space. Experiments on MTEB show that SpecMRL outperforms MRL at all evaluated reduced dimensions, with larger gains under more aggressive compression. These results suggest that the learned spectral basis better prioritizes informative directions, enabling more effective representation compression. Notably, SpecMRL extends the performance frontier of MRL, surpassing full-dimensional MRL with only half the embedding dimensions.
PDF arXiv Code
arXiv cs.AICodeOct 16▲ 7
A Conditional No-Hair Theorem for Transformers: How Language Models Forget Their Prompt
Shuyang Xiang
We ask how much an autoregressive transformer retains of its prompt after generating $t$ tokens. Consider two different prompts followed by the same continuation text, so that any difference in their hidden states comes from the prompt alone, and we find that the distance between their hidden states decays approximately as a power law $t^{-α}$, where $α=1$ is the rate at which a fixed prompt's share of uniform attention shrinks and smaller values mean longer memory. Trained models forget fast ($α\approx0.55$--$0.7$), untrained ones slowly ($α\approx0.2$--$0.5$). We prove that this offset is $O\big(P(\log t)^{L-1}/t\big)$ for a prompt of length $P$ and depth $L$. The bound holds for finite-depth pre-LN softmax transformers with bounded attention logits, under one additional condition: the pre-norm activations stay away from zero. The mechanism is geometric: the prompt's direct influence is diluted as the context grows, and \emph{relay} --- continuation tokens passing on what earlier tokens read from the prompt --- can only add logarithmic factors. This is what we call the \emph{conditional no-hair theorem}: given the text generated so far, the prompt's influence on the next token…
PDF arXiv Code
arXiv cs.LGCodeOct 15▲ 106
Behavioral Compositionality of Sparse Autoencoder Representations
Haotian Ma, Ziye Mei, Frederic Sala
Sparse autoencoders (SAEs) represent model activations with sparse, linearly decoded features, but the downstream effects of those features need not compose additively. We introduce SAE Additivity Residuals (SAR), which measure the behavioral variation left unexplained above a chosen interaction order. Under independent feature subset sampling, population SAR is exactly the normalized energy of interactions above that order and bounds coalition ordering error for the optimal low order approximation. To distinguish compositionality associated with the SAE representation from that induced by downstream computation, we compare learned SAE coordinates with sum preserving random rotations of the same selected subspace and find lower first order residuals for the learned coordinates across all 18 model and behavior settings. The coordinates are nevertheless not fully additive: pairwise terms improve held out coalition ordering in every setting and outperform parameter matched models using randomly selected third order terms. In a combined steering experiment, pairwise modeling raises rank fidelity from 0.5874 to 0.7790 and reduces ordering error from 0.1444 to 0.0606. Together, these…
PDF arXiv Code
arXiv cs.LGCodeOct 14▲ 5
Your Transformer Is Secretly Elastic
Walter Simoncini, Dheeraj Varghese, Gertjan J. Burghouts, Cees G. M. Snoek
Scaling up transformers has produced models with broad capabilities, deployed across heterogeneous hardware and fluctuating workloads. Yet, model families offer only a few fixed sizes (e.g., S, B, L), forcing users to commit to a model based on hardware and latency constraints. This rigidity is often suboptimal under varying query loads. We introduce a retraining-free method that elastifies pre-trained transformers into a single model that adapts on-the-fly across a continuum of computational budgets. Our key contribution is a differentiable relaxation of the functional unit-ranking problem: unlike prior work, which optimizes a single importance weight per MLP block or attention head, we rank every individual MLP neuron and attention head jointly, on a common scale, by minimizing the model's loss under soft pruning across many sparsity targets, all at once. Across 27 vision and text transformer encoders ranging from 22M to 8B parameters, our method yields smooth degradation up to 60% sparsity and outperforms one-shot baselines under deep pruning. For example, pruning DINOv3 ViT-H+/16 to 50% sparsity removes 427M parameters while reducing ImageNet-1k linear-probing accuracy by…
PDF arXiv Code
arXiv cs.LGCodeOct 14▲ 9
Continuum Attention
Afshin Khadangi
We introduce Continuum Attention (CA), a causal Transformer attention operator that replaces discrete key and value lookup with integration over a continuous, piecewise linear memory curve. Each query induces an exponential relevance density along the curve. Integrating the density and interpolated values yields exact normalized attention from closed form exponential moments, including half token boundary contributions. The gradients follow analytically from moments of order at most two, enabling parallel GPU implementations while retaining quadratic sequence complexity and the standard Transformer decoder. Compared with ordinary softmax and focal attention, CA reduces mean validation negative log likelihood by 0.0200 and 0.0480 nats per byte, respectively, winning all paired comparisons. An additional disjoint validation slice yields a 0.0457 nats per byte advantage over the strongest selected softmax baseline, with a paired 95% confidence interval. These findings motivate continuous integration as a promising attention alternative while leaving larger scale generalization open.
PDF arXiv Code
arXiv cs.LGCodeOct 10▲ 3
ShellElastic: Scalable Elastic Pretraining with Simultaneous Maximal Updates and Hyperparameter Transfer
Can Jin, Yan Kang, Yifan Gong, Chongjian Ge, Mingze Xu +3
Foundation models need to support diverse inference budgets, yet training a separate model for each budget duplicates pretraining effort. Elastic models train several nested sub-networks inside one weight-shared supernet, but existing recipes face three obstacles: variable per-step workloads that hinder distributed scaling, limited hyperparameter (HP) transfer across model sizes, and no guarantee that all sub-networks are maximally updated at once. We present \Ours, an elastic transformer that organizes hidden dimensions into a dense core and successive shells, so that every hidden matrix is block lower triangular. One full-width forward pass contains every sub-network's exact hidden states as prefixes, so joint training keeps a fixed computation graph and costs about as much per step as training the widest sub-network alone. We further introduce \muE, an extension of maximal update parameterization (\mup) to \Ours and other elastic architectures. Its scaling rules provably keep all sub-networks of a supernet maximally updated simultaneously (\muI), and keep doing so when the supernet is widened or given more shells; sweeps confirm that HPs tuned on a small non-elastic proxy stay…
PDF arXiv Code

Papers

Topics in this field

graph learning · graph neural · graph neural networks256 papers41 this week
gradient descent · learning rate · neural networks209 papers35 this week
neural networks · neural network · arnold networks143 papers33 this week
memory efficient · optimizers · low rank adaptation151 papers24 this week
bit quantization · quantization aware training · quantization aware147 papers16 this week
linear attention · softmax attention · transformer attention104 papers13 this week

Active contributors · 90 days

★ Spotlight this month

arXiv cs.LGCodeOct 14▲ 5

Your Transformer Is Secretly Elastic

Walter Simoncini, Dheeraj Varghese, Gertjan J. Burghouts, Cees G. M. Snoek

Scaling up transformers has produced models with broad capabilities, deployed across heterogeneous hardware and fluctuating workloads. Yet, model families offer only a few fixed sizes (e.g., S, B, L), forcing users to commit to a model based on hardware and latency constraints. This rigidity is often suboptimal under varying query loads. We introduce a retraining-free method that elastifies pre-trained transformers into a single model that adapts on-the-fly across a continuum of computational budgets. Our key contribution is a differentiable relaxation of the functional unit-ranking problem: unlike prior work, which optimizes a single importance weight per MLP block or attention head, we rank every individual MLP neuron and attention head jointly, on a common scale, by minimizing the model's loss under soft pruning across many sparsity targets, all at once. Across 27 vision and text transformer encoders ranging from 22M to 8B parameters, our method yields smooth degradation up to 60% sparsity and outperforms one-shot baselines under deep pruning. For example, pruning DINOv3 ViT-H+/16 to 50% sparsity removes 427M parameters while reducing ImageNet-1k linear-probing accuracy by…

PDF arXiv page Details & code

Activity · 12 weeks

Coming with more sources: repos

Open problems in Graph learning

Quantization error propagation in recurrent states1 paper raises it · 1 addresses it
Low-precision recurrent states in linear attention degrade accuracy due to error accumulation over time and across dimensions.
- STEPQuant: When and Where Errors Matter in Delta-Rule Recurrent State Quantizationraises · addresses
Train-rollout discrepancy in low-precision RL1 paper raises it · 1 addresses it
Quantization in RL training and rollout paths can diverge, reducing the effectiveness of low-precision methods.
- TRACE: Rollout-Guided Quantization-Aware Training for FP4 Reinforcement Learning of MoE Language Modelsraises · addresses
Distribution shift in pruning for decoding1 paper raises it · 1 addresses it
Pruning guided by natural sequences fails to account for self-generated token distributions during decoding, hurting performance.
- SparseDecoding: Decoding-Aware Pruning for Accurate and Efficient LLM Inferenceraises · addresses
Lack of benchmarks for graph decision-making in System One models1 paper raises it · 1 addresses it
Existing benchmarks do not adequately evaluate System One models on graph tasks, limiting progress in this area.
- GraphDecisionBench: Benchmarking System One Models on Graph Tasksraises · addresses
Calibration dependency in diffusion model quantization1 paper raises it · 1 addresses it
Post-training quantization for diffusion models often relies on calibration data, which is impractical due to shifting activation distributions.
- SoloQ: Calibration-Free Quantization for Diffusion Language Modelsraises · addresses

Named by an AI from these papers' abstracts.

Fields · AI · updated 48m ago

Graph learning

Found by the topic model: graph learning, gradient descent, bit quantization.

162papers this week−29.3% vs last week

All papers · 1k With code · 11

Newest Most relevant

981 papers this month

Recent papers

All in Research →

arXiv cs.CLCodeOct 19▲ 79
SpecMRL: Redesign Matryoshka Representation Learning within Parametric Spectrum
Zhongxin Chen, Songhao Wu, Heng Cui, Cong Li, Xiao Zhou +1
Matryoshka Representation Learning (MRL) is designed to support flexible representation sizes by jointly optimizing nested prefixes. While effective, this design lacks an explicit constraint to ensure that the leading portion provides a suitable compact encoding. To address this, we propose SpecMRL, which reorganizes the feature space to enable more effective prefix readout. Specifically, we extend MRL with a parameter-derived spectral basis organized with feature importance. SpecMRL otherwise follows standard MRL, with prefix readout performed in the learned spectral space. Experiments on MTEB show that SpecMRL outperforms MRL at all evaluated reduced dimensions, with larger gains under more aggressive compression. These results suggest that the learned spectral basis better prioritizes informative directions, enabling more effective representation compression. Notably, SpecMRL extends the performance frontier of MRL, surpassing full-dimensional MRL with only half the embedding dimensions.
PDF arXiv Code
arXiv cs.AICodeOct 16▲ 7
A Conditional No-Hair Theorem for Transformers: How Language Models Forget Their Prompt
Shuyang Xiang
We ask how much an autoregressive transformer retains of its prompt after generating $t$ tokens. Consider two different prompts followed by the same continuation text, so that any difference in their hidden states comes from the prompt alone, and we find that the distance between their hidden states decays approximately as a power law $t^{-α}$, where $α=1$ is the rate at which a fixed prompt's share of uniform attention shrinks and smaller values mean longer memory. Trained models forget fast ($α\approx0.55$--$0.7$), untrained ones slowly ($α\approx0.2$--$0.5$). We prove that this offset is $O\big(P(\log t)^{L-1}/t\big)$ for a prompt of length $P$ and depth $L$. The bound holds for finite-depth pre-LN softmax transformers with bounded attention logits, under one additional condition: the pre-norm activations stay away from zero. The mechanism is geometric: the prompt's direct influence is diluted as the context grows, and \emph{relay} --- continuation tokens passing on what earlier tokens read from the prompt --- can only add logarithmic factors. This is what we call the \emph{conditional no-hair theorem}: given the text generated so far, the prompt's influence on the next token…
PDF arXiv Code
arXiv cs.LGCodeOct 15▲ 106
Behavioral Compositionality of Sparse Autoencoder Representations
Haotian Ma, Ziye Mei, Frederic Sala
Sparse autoencoders (SAEs) represent model activations with sparse, linearly decoded features, but the downstream effects of those features need not compose additively. We introduce SAE Additivity Residuals (SAR), which measure the behavioral variation left unexplained above a chosen interaction order. Under independent feature subset sampling, population SAR is exactly the normalized energy of interactions above that order and bounds coalition ordering error for the optimal low order approximation. To distinguish compositionality associated with the SAE representation from that induced by downstream computation, we compare learned SAE coordinates with sum preserving random rotations of the same selected subspace and find lower first order residuals for the learned coordinates across all 18 model and behavior settings. The coordinates are nevertheless not fully additive: pairwise terms improve held out coalition ordering in every setting and outperform parameter matched models using randomly selected third order terms. In a combined steering experiment, pairwise modeling raises rank fidelity from 0.5874 to 0.7790 and reduces ordering error from 0.1444 to 0.0606. Together, these…
PDF arXiv Code
arXiv cs.LGCodeOct 14▲ 5
Your Transformer Is Secretly Elastic
Walter Simoncini, Dheeraj Varghese, Gertjan J. Burghouts, Cees G. M. Snoek
Scaling up transformers has produced models with broad capabilities, deployed across heterogeneous hardware and fluctuating workloads. Yet, model families offer only a few fixed sizes (e.g., S, B, L), forcing users to commit to a model based on hardware and latency constraints. This rigidity is often suboptimal under varying query loads. We introduce a retraining-free method that elastifies pre-trained transformers into a single model that adapts on-the-fly across a continuum of computational budgets. Our key contribution is a differentiable relaxation of the functional unit-ranking problem: unlike prior work, which optimizes a single importance weight per MLP block or attention head, we rank every individual MLP neuron and attention head jointly, on a common scale, by minimizing the model's loss under soft pruning across many sparsity targets, all at once. Across 27 vision and text transformer encoders ranging from 22M to 8B parameters, our method yields smooth degradation up to 60% sparsity and outperforms one-shot baselines under deep pruning. For example, pruning DINOv3 ViT-H+/16 to 50% sparsity removes 427M parameters while reducing ImageNet-1k linear-probing accuracy by…
PDF arXiv Code
arXiv cs.LGCodeOct 14▲ 9
Continuum Attention
Afshin Khadangi
We introduce Continuum Attention (CA), a causal Transformer attention operator that replaces discrete key and value lookup with integration over a continuous, piecewise linear memory curve. Each query induces an exponential relevance density along the curve. Integrating the density and interpolated values yields exact normalized attention from closed form exponential moments, including half token boundary contributions. The gradients follow analytically from moments of order at most two, enabling parallel GPU implementations while retaining quadratic sequence complexity and the standard Transformer decoder. Compared with ordinary softmax and focal attention, CA reduces mean validation negative log likelihood by 0.0200 and 0.0480 nats per byte, respectively, winning all paired comparisons. An additional disjoint validation slice yields a 0.0457 nats per byte advantage over the strongest selected softmax baseline, with a paired 95% confidence interval. These findings motivate continuous integration as a promising attention alternative while leaving larger scale generalization open.
PDF arXiv Code
arXiv cs.LGCodeOct 10▲ 3
ShellElastic: Scalable Elastic Pretraining with Simultaneous Maximal Updates and Hyperparameter Transfer
Can Jin, Yan Kang, Yifan Gong, Chongjian Ge, Mingze Xu +3
Foundation models need to support diverse inference budgets, yet training a separate model for each budget duplicates pretraining effort. Elastic models train several nested sub-networks inside one weight-shared supernet, but existing recipes face three obstacles: variable per-step workloads that hinder distributed scaling, limited hyperparameter (HP) transfer across model sizes, and no guarantee that all sub-networks are maximally updated at once. We present \Ours, an elastic transformer that organizes hidden dimensions into a dense core and successive shells, so that every hidden matrix is block lower triangular. One full-width forward pass contains every sub-network's exact hidden states as prefixes, so joint training keeps a fixed computation graph and costs about as much per step as training the widest sub-network alone. We further introduce \muE, an extension of maximal update parameterization (\mup) to \Ours and other elastic architectures. Its scaling rules provably keep all sub-networks of a supernet maximally updated simultaneously (\muI), and keep doing so when the supernet is widened or given more shells; sweeps confirm that HPs tuned on a small non-elastic proxy stay…
PDF arXiv Code