CapRL++ applies reinforcement learning with verifiable rewards to dense image and video captioning by scoring captions via the accuracy of a vision-free LLM answering MCQs from the caption alone.
hub Mixed citations
SemDeDup: Data-efficient learning at web-scale through semantic deduplication
Mixed citation behavior. Most common role is background (55%).
30
Pith papers citing it
Background
55% of classified citations
abstract
Progress in machine learning has been driven in large part by massive increases in data. However, large web-scale datasets such as LAION are largely uncurated beyond searches for exact duplicates, potentially leaving much redundancy. Here, we introduce SemDeDup, a method which leverages embeddings from pre-trained models to identify and remove semantic duplicates: data pairs which are semantically similar, but not exactly identical. Removing semantic duplicates preserves performance and speeds up learning. Analyzing a subset of LAION, we show that SemDeDup can remove 50% of the data with minimal performance loss, effectively halving training time. Moreover, performance increases out of distribution. Also, analyzing language models trained on C4, a partially curated dataset, we show that SemDeDup improves over prior approaches while providing efficiency gains. SemDeDup provides an example of how simple ways of leveraging quality embeddings can be used to make models learn faster with less data.
hub tools
citation-role summary
background 7
method 3
dataset 1
citation-polarity summary
representative citing papers
RAD uses HNSW approximate nearest neighbor search augmented with a bitmap for Jaccard similarity to deliver 0.94-0.97 recall and up to 8x throughput versus LSH on evolving LLM-scale datasets up to 30M documents.
AutoSelection discovers data recipes from a 90K instruction pool that outperform full-data training and other selectors on reasoning tasks for SFT across multiple models.
OmniShotCut treats shot boundary detection as structured relational prediction via a shot-query Transformer, uses fully synthetic transitions for training data, and releases OmniShotCutBench for evaluation.
On-policy self-distillation with sampled demonstrations reduces rollout diversity by amplifying existing probability gaps in the base model, unlike ideal RL which preserves ratios among correct outputs.
Repetition of training data produces a systematic eval loss peak at intermediate repeat counts whose location scales with model size, quantifiable as large compute-equivalent loss even at modest repetition fractions.
HullFT performs test-time finetuning by sparse convex reconstruction of query embeddings followed by gradient caching on repeated examples, yielding better quality-efficiency tradeoffs than prior TTFT methods.
Learning-Zone Energy is a new online data selection framework for RL post-training that retains 40% of data per step yet matches or exceeds full-data baselines on math tasks with 36% lower FLOPs.
SEED models data selection as Weighted Independent Set on a similarity graph, using node value calibration and local scale normalization to produce compact high-quality training subsets that outperform prior methods on instruction tuning and segmentation tasks.
Data curation alone raises VLM accuracy by more than 11 points on average across many benchmarks while cutting required training compute by up to 87 times.
A video transfer pipeline augments simulated VLA data into realistic videos while preserving actions, yielding consistent performance gains on robot benchmarks such as 8% on Robotwin 2.0.
Dynamic per-epoch cluster scaling that preserves relative cluster order and upsamples the long tail cuts VLM pre-training cost and improves long-tail zero-shot accuracy.
Influence scoring can use only forward passes: CountSketch-compressed outer products of the LM-head residual and final hidden state give accurate attribution and valuation from 14M to 32B parameters.
MOSAIC is a scaling-aware data selection framework that outperforms baselines in training end-to-end autonomous driving planners, achieving comparable or better EPDMS scores with up to 80% less data.
COMPACT synthesizes compositional visual instruction data to reduce VIT training data by 90% while achieving 100.2% of full performance across eight multimodal benchmarks.
DCLM-Baseline dataset lets a 7B model reach 64% 5-shot MMLU accuracy after 2.6T tokens, beating prior open-data models by 6.6 points on MMLU with 40% less compute.
MetaCLIP curates balanced 400M-pair subsets from CommonCrawl that outperform CLIP data, reaching 70.8% zero-shot ImageNet accuracy on ViT-B versus CLIP's 68.3%.
DepTrans uses reinforcement-aligned multi-task training and dependency-guided iterative refinement to raise repository-level C-to-Rust compilation success to 60.7% and computational accuracy to 43.5%.
SkillCenter provides 216,938 source-grounded skills across 24 domains via an automated LLM-based pipeline, with a downstream evaluation showing skills help agents only when the task exceeds the model's knowledge and retrieval surfaces the right skill.
An iterative bootstrapped self-filtering approach selects balanced clean and diverse subsets from noisy vision-language datasets to train improved CLIP models.
LCC-LLM creates a code-centric dataset and RAG-based LLM framework that reaches 0.634 average semantic similarity on 43 malware tasks and 10/10 pass rate in real-world case studies.
NVILA improves on VILA with a scale-then-compress visual token strategy and full-lifecycle efficiency optimizations, matching or exceeding leading VLMs on image and video benchmarks while reducing training cost 1.9-5.1x and latencies 1.2-2.8x.
The paper surveys hallucination in LLMs with an innovative taxonomy, factors, detection methods, benchmarks, mitigation strategies, and open research directions.
Merlin achieves byte-exact deduplication of text at up to 8.7 GB/s using SIMD-optimized hashing, reducing LLM context sizes by 13.9-71% with no data loss.
citing papers explorer
-
CapRL++: Unified Reinforcement Learning with Verifiable Rewards for Dense Image and Video Captioning
CapRL++ applies reinforcement learning with verifiable rewards to dense image and video captioning by scoring captions via the accuracy of a vision-free LLM answering MCQs from the caption alone.
-
FOLD: Fuzzy Online Deduplication for Very Large Evolving Datasets via Approximate Nearest Neighbor Search
RAD uses HNSW approximate nearest neighbor search augmented with a bitmap for Jaccard similarity to deliver 0.94-0.97 recall and up to 8x throughput versus LSH on evolving LLM-scale datasets up to 30M documents.
-
From Instance Selection to Fixed-Pool Data Recipe Search for Supervised Fine-Tuning
AutoSelection discovers data recipes from a 90K instruction pool that outperform full-data training and other selectors on reasoning tasks for SFT across multiple models.
-
OmniShotCut: Holistic Relational Shot Boundary Detection with Shot-Query Transformer
OmniShotCut treats shot boundary detection as structured relational prediction via a shot-query Transformer, uses fully synthetic transitions for training data, and releases OmniShotCutBench for evaluation.
-
On-Policy Self-Distillation with Sampled Demonstrations Reduces Output Diversity
On-policy self-distillation with sampled demonstrations reduces rollout diversity by amplifying existing probability gaps in the base model, unlike ideal RL which preserves ratios among correct outputs.
-
Internal Data Repetition Destroys Language Models
Repetition of training data produces a systematic eval loss peak at intermediate repeat counts whose location scales with model size, quantifiable as large compute-equivalent loss even at modest repetition fractions.
-
Efficient Test-Time Finetuning of LLMs via Convex Reconstruction and Gradient Caching
HullFT performs test-time finetuning by sparse convex reconstruction of query embeddings followed by gradient caching on repeated examples, yielding better quality-efficiency tradeoffs than prior TTFT methods.
-
Learning-Zone Energy: Online Data Selection for Efficient RL Post-Training
Learning-Zone Energy is a new online data selection framework for RL post-training that retains 40% of data per step yet matches or exceeds full-data baselines on math tasks with 36% lower FLOPs.
-
SEED: Targeted Data Selection by Weighted Independent Set
SEED models data selection as Weighted Independent Set on a similarity graph, using node value calibration and local scale normalization to produce compact high-quality training subsets that outperform prior methods on instruction tuning and segmentation tasks.
-
20/20 Vision Language Models: A Prescription for Better VLMs through Data Curation Alone
Data curation alone raises VLM accuracy by more than 11 points on average across many benchmarks while cutting required training compute by up to 87 times.
-
Seeing Realism from Simulation: Efficient Video Transfer for Vision-Language-Action Data Augmentation
A video transfer pipeline augments simulated VLA data into realistic videos while preserving actions, yielding consistent performance gains on robot benchmarks such as 8% on Robotwin 2.0.
-
Dynamic Cluster Data Sampling for Efficient and Long-Tail-Aware Vision-Language Pre-training
Dynamic per-epoch cluster scaling that preserves relative cluster order and upsamples the long tail cuts VLM pre-training cost and improves long-tail zero-shot accuracy.
-
Sketching the Readout of Large Language Models for Scalable Data Attribution and Valuation
Influence scoring can use only forward passes: CountSketch-compressed outer products of the LM-head residual and final hidden state give accurate attribution and valuation from 14M to 32B parameters.
-
Scaling-Aware Data Selection for End-to-End Autonomous Driving Systems
MOSAIC is a scaling-aware data selection framework that outperforms baselines in training end-to-end autonomous driving planners, achieving comparable or better EPDMS scores with up to 80% less data.
-
Visual Compositional Tuning
COMPACT synthesizes compositional visual instruction data to reduce VIT training data by 90% while achieving 100.2% of full performance across eight multimodal benchmarks.
-
DataComp-LM: In search of the next generation of training sets for language models
DCLM-Baseline dataset lets a 7B model reach 64% 5-shot MMLU accuracy after 2.6T tokens, beating prior open-data models by 6.6 points on MMLU with 40% less compute.
-
Demystifying CLIP Data
MetaCLIP curates balanced 400M-pair subsets from CommonCrawl that outperform CLIP data, reaching 70.8% zero-shot ImageNet accuracy on ViT-B versus CLIP's 68.3%.
-
Dependency-Guided Repository-Level C-to-Rust Translation with Reinforcement Alignment
DepTrans uses reinforcement-aligned multi-task training and dependency-guided iterative refinement to raise repository-level C-to-Rust compilation success to 60.7% and computational accuracy to 43.5%.
-
SkillCenter: A Large-Scale Source-Grounded Skill Library for Autonomous AI Agents
SkillCenter provides 216,938 source-grounded skills across 24 domains via an automated LLM-based pipeline, with a downstream evaluation showing skills help agents only when the task exceeds the model's knowledge and retrieval surfaces the right skill.
-
Data Selection Through Iterative Self-Filtering for Vision-Language Settings
An iterative bootstrapped self-filtering approach selects balanced clean and diverse subsets from noisy vision-language datasets to train improved CLIP models.
-
LCC-LLM: Leveraging Code-Centric Large Language Models for Malware Attribution
LCC-LLM creates a code-centric dataset and RAG-based LLM framework that reaches 0.634 average semantic similarity on 43 malware tasks and 10/10 pass rate in real-world case studies.
-
NVILA: Efficient Frontier Visual Language Models
NVILA improves on VILA with a scale-then-compress visual token strategy and full-lifecycle efficiency optimizations, matching or exceeding leading VLMs on image and video benchmarks while reducing training cost 1.9-5.1x and latencies 1.2-2.8x.
-
A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions
The paper surveys hallucination in LLMs with an innovative taxonomy, factors, detection methods, benchmarks, mitigation strategies, and open research directions.
-
Merlin: Deterministic Byte-Exact Deduplication for Lossless Context Optimization in Large Language Model Inference
Merlin achieves byte-exact deduplication of text at up to 8.7 GB/s using SIMD-optimized hashing, reducing LLM context sizes by 13.9-71% with no data loss.
-
Byte-Exact Deduplication in Retrieval-Augmented Generation: A Three-Regime Empirical Analysis Across Public Benchmarks
Byte-exact deduplication reduces RAG context size by 0.16% to 80.34% across three regimes with zero measurable quality regression per multi-vendor LLM evaluation.
-
GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models
GLM-4.5, a 355B-parameter MoE model with hybrid reasoning, scores 70.1% on TAU-Bench, 91.0% on AIME 24, and 64.2% on SWE-bench Verified while ranking 3rd overall and 2nd on agentic benchmarks.
-
Towards EnergyGPT: A Large Language Model Specialized for the Energy Sector
Fine-tuned LLaMA 3.1-8B variants for the energy sector outperform the base model on domain QA benchmarks, with LoRA delivering similar gains at lower training cost.
-
Cosmos World Foundation Model Platform for Physical AI
The Cosmos platform supplies open-source pre-trained world models and supporting tools for building fine-tunable digital world simulations to train Physical AI.
- DataComp-VLM: Improved Open Datasets for Vision-Language Models
- Spectral Certificates and Projection-DPP Rounding for Determinantal MAP Selection