CORE: Improving Compositional Reasoning in MLLM Embedding via Reranker Distillation

• URL: http://arxiv.org/abs/2609.04083v1
• Date: Thu, 03 Sep 2026 16:50:29 GMT
• Title: CORE: Improving Compositional Reasoning in MLLM Embedding via Reranker Distillation
• Abstract summary: We distill the compositional judgments of a cross-attentive reranker into an embedding model.<n>We show that CORE-RERANKER-8B achieves an 82.7% total average, outperforming Jina-Reranker by 10.7 points.<n>The improvements transfer to the MCMR benchmark without sacrificing retrieval performance on COCO and Flickr30K.
• Score: 49.0564416580726
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: MLLM-based embedding models remain limited in compositional retrieval, often failing to distinguish scenes containing the same concepts but different attribute-object bindings. Yet the same backbone can resolve such distinctions when used as a cross-attentive reranker, motivating us to distill its compositional judgments into the embedding model. We propose CORE, which synthesizes candidate lists spanning five compositional matching levels and introduces a Rank-KL objective that trains the embedding model to reproduce the reranker's fine-grained ranking. We further introduce a graded evaluation protocol and compare contrastive learning, pairwise CoSENT, and listwise Rank-KL under the same data and tuning budget. Our comparison shows that both CoSENT and Rank-KL use the multi-level supervision more effectively than contrastive learning, with Rank-KL achieving the strongest overall performance. Across three compositional reasoning benchmarks (COLA, SUGARCREPE++, NEGBENCH), CORE-RERANKER-8B achieves an 82.7% total average, outperforming Jina-Reranker by 10.7 points, while CORE-EMBED-8B achieves the best total average (0.666) among all evaluated embedding models. The improvements transfer to the MCMR benchmark without sacrificing retrieval performance on COCO and Flickr30K.

Related papers

• DataFlex-RL: An Evaluation Platform for RLVR Data Policies [11.073211317719393]
  Data policies for reinforcement learning with verifiable rewards determine which rollouts are used, how strongly they are weighted, and which domains contribute to subsequent training batches.<n>We introduce DataFlex-RL, an evaluation platform for comparing these choices under a common GRPO recipe.<n>Our primary experiment evaluates 13 configurations across 12 matched seeds using Qwen2.5-7B-Base and 12 mathematics, logic, and science benchmarks.
  arXiv  Detail & Related papers  (2026-09-05T14:04:50Z)
• On-Policy Distillation Meets Off-Policy GRPO: Training Compact Instruction-Following Rerankers [0.0]
  We revisit reranker distillation through the lens of reinforcement learning.<n>We propose a two-stage framework combining off-policy teacher optimization with on-policy student distillation.
  arXiv  Detail & Related papers  (2026-09-01T23:36:19Z)
• Rethinking Multi-Branch and Cross-Backbone Fusion for Vehicle Re-Identification in the Foundation-Model Era [22.36803510099264]
  Multi-branch architectures and CNN-Transformer fusion have long been regarded as effective ways to improve vehicle re-identification (Re-ID)<n>A single DINOv3-pretrained ConvNeXt trained with a tuned recipe achieves 88.19 mAP on VeRi-Wild Small and 77.47 mAP on VeRi-Wild Large.<n>We further study cross-backbone fusion using an asymmetric frozen-anchor strategy to combine ConvNeXt and Vision Transformer representations.
  arXiv  Detail & Related papers  (2026-07-24T08:10:53Z)
• Similarity-Guided Curriculum Fine-Tuning of LLMs for Neural Architecture Synthesis [48.83701310501069]
  We introduce a MinHash-based similarity scheduling framework that constructs a progressive curriculum over neural architecture code for neural architecture search (NAS)<n>We evaluate the framework on OlympicCoder-7B within the LEMUR benchmark on CIFAR-10 image classification.<n>We report a cross-dataset transfer observation on SVHN, where direct base-model generation without curriculum yields 27% peak SR at substantially lower accuracy (60.5%) than the CIFAR-10 equivalent.
  arXiv  Detail & Related papers  (2026-07-13T14:14:30Z)
• Putting HUMANS first: Efficient LAM Evaluation with Human Preference Alignment [53.72927532626824]
  We show that subsets of just 50 examples (0.3% of data) can achieve over 0.93 Pearson correlation with full benchmark scores.<n>To better predict preferences, we trained regression models on these selected subsets, achieving 0.98 correlation.<n>This demonstrates that in regression modeling, well-curated subsets outpredict the full benchmark, showing quality over quantity.
  arXiv  Detail & Related papers  (2026-04-20T00:57:31Z)
• COGNAC at SemEval-2026 Task 5: LLM Ensembles for Human-Level Word Sense Plausibility Rating in Challenging Narratives [0.0]
  We describe our system for rating the plausibility of given word senses of homonyms in short stories on a 5-point Likert scale.<n>System is evaluated by the unweighted average of accuracy (within one standard deviation of mean human judgments) and Spearman Rank Correlation.
  arXiv  Detail & Related papers  (2026-03-16T20:42:02Z)
• SAGE: A Realistic Benchmark for Semantic Understanding [9.688555356614044]
  We introduce SAGE (Semantic Alignment & Generalization Evaluation), a rigorous benchmark designed to assess both embedding models and similarity metrics.<n>Our comprehensive evaluation of 9 embedding models and classical metrics reveals significant performance gaps.<n>OpenAI's text-embedding-3-small achieves the highest clustering performance (0.483) but demonstrates extreme brittleness with the lowest robustness score (0.011)
  arXiv  Detail & Related papers  (2025-09-25T15:27:15Z)
• Variational Supervised Contrastive Learning [50.79938854370321]
  We propose Variational Supervised Contrastive Learning (VarCon), which reformulates supervised contrastive learning as variational inference over latent class variables.<n>VarCon achieves state-of-the-art performance for contrastive learning frameworks, reaching 79.36% Top-1 accuracy on ImageNet-1K and 78.29% on CIFAR-100 with a ResNet-50 encoder.
  arXiv  Detail & Related papers  (2025-06-09T04:19:12Z)
• Dr. SoW: Density Ratio of Strong-over-weak LLMs for Reducing the Cost of Human Annotation in Preference Tuning [15.776175440446414]
  We introduce Dr.SoW (Density Ratio of Strong over Weak) a cost-effective method that eliminates the reliance for human annotation.<n>Dr.SoW uses the log-density ratio between a better-aligned and a less-aligned LLM as a reward signal.<n>We preference-tune Llama-3-8B-Instruct using data annotated by Dr.SoW.
  arXiv  Detail & Related papers  (2024-11-04T18:54:39Z)
• Language Model Preference Evaluation with Multiple Weak Evaluators [89.90733463933431]
  We introduce PGED, a novel approach that leverages multiple model-based evaluators to construct preference graphs, and then ensembles and denoises these graphs for acyclic, non-contradictory evaluation results.<n>We demonstrate PGED's superiority in three applications: 1) model ranking for evaluation, 2) response selection for test-time scaling, and 3) data selection for model fine-tuning.
  arXiv  Detail & Related papers  (2024-10-14T01:57:25Z)
• COLO: A Contrastive Learning based Re-ranking Framework for One-Stage Summarization [84.70895015194188]
  We propose a Contrastive Learning based re-ranking framework for one-stage summarization called COLO. COLO boosts the extractive and abstractive results of one-stage systems on CNN/DailyMail benchmark to 44.58 and 46.33 ROUGE-1 score.
  arXiv  Detail & Related papers  (2022-09-29T06:11:21Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.