{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.00622", "kind": "paper", "url": "https://arxiv.org/abs/2606.00622"}, {"kind": "website", "url": "https://frenkie-chiang.github.io/MM-Snowball"}], "authors": ["Yue Jiang", "Xue Jiang", "Lihua Zhang", "Zhiqiang Wang", "Yuhang Lu", "Peng Wang", "Bo Han", "Feng Zheng", "Dingkang Yang"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.00622", "source_url": "https://arxiv.org/abs/2606.00622"}, "categories": ["Cybersecurity", "Multimodal"], "description": {"en": "MM-Snowball is a benchmark for diagnosing hallucination snowballing in multimodal multi-turn dialogue, with fine-grained analysis. It includes data and code via a project page."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.00622", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.00622", "title": "MM-Snowball"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Fudan University", "Hong Kong Baptist University", "Southern University of Science and Technology", "Hong Kong University of Science and Technology", "Huawei Technologies (China)", "Cyber University of Korea", "University of Hong Kong"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.00622", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.00622", "source_url": "https://openalex.org/W7163172548"}, "key": "claire-radar:2606.00622", "modality": null, "name": "MM-Snowball", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "High", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "unknown", "origin_record_id": "bm_mm-snowball_ce633f35", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "fbda0dd6b54cc3dac867e5c8cac28780d422078762958a3e3f528286045331d2", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.00622"}, "publisher": null, "released": "2026-05-30", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.00622", "source_url": "https://arxiv.org/abs/2606.00622"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-00622", "source": "claire_radar", "source_benchmark_id": "2606.00622", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["Cybersecurity"], "area": "Multimodal", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": null, "githubScope": null, "githubStars": null, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": null, "hfPaperUrl": "https://huggingface.co/papers/2606.00622", "observedAt": "2026-10-04T03:02:42.754940Z", "signalStatus": {"githubStars": {"state": "not_applicable"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "state": "unavailable"}}, "status": "unavailable"}, "capabilities": [], "confidence": "High", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "34a3475b761d9810c0eb1b66e2adde6c1559b2de348d06a491b1b222eed3cdfe", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "MM-Snowball is a benchmark for diagnosing hallucination snowballing in multimodal multi-turn dialogue, with fine-grained analysis. It includes data and code via a project page.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence", "public artifact URL"], "snippet": "To address this, we introduce MM-Snowball, the first benchmark for fine-grained diagnosis of hallucination snowballing within dialogues."}, "familyId": "bmf_17f078aecdcc", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_mm-snowball_ce633f35", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Cybersecurity"], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2606.00622", "paper": "https://arxiv.org/abs/2606.00622", "pdf": "https://arxiv.org/pdf/2606.00622", "project": "https://frenkie-chiang.github.io/MM-Snowball", "report": "https://arxiv.org/abs/2606.00622"}, "metrics": [], "motivation": "Multimodal large language models (MLLMs) demonstrate remarkable visual understanding, yet their reliability in interactive settings is severely undermined by hallucination snowballing: a phenomenon where initial errors amplify across conversational turns, leading to a collapse in coherence.", "name": "MM-Snowball", "oneLine": "MM-Snowball is a benchmark for diagnosing hallucination snowballing in multimodal multi-turn dialogue, with fine-grained analysis. It includes data and code via a project page.", "paperTitle": "MM-Snowball: Evaluating and Mitigating Hallucination Snowballing in Multimodal Multi-Turn Dialogue", "primaryDomain": "Cybersecurity", "publication": {"evidence": "Accepted by The International Conference on Machine Learning (ICML 2026)", "evidenceLevel": "author-claim", "evidenceUrl": "https://arxiv.org/abs/2606.00622", "source": "arxiv-comments", "status": "acceptance_claimed", "venue": "International Conference on Machine Learning (ICML 2026)", "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {}, "readiness": "Inspectable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-05-30", "source": {"authors": ["Yue Jiang and Xue Jiang and Lihua Zhang and Zhiqiang Wang and Yuhang Lu and Peng Wang and Bo Han and Feng Zheng and Dingkang Yang"], "categories": ["cs.CV"], "id": "2606.00622", "title": "MM-Snowball: Evaluating and Mitigating Hallucination Snowballing in Multimodal Multi-Turn Dialogue", "type": "arxiv", "url": "https://arxiv.org/abs/2606.00622"}, "sourceUpdatedAt": "2026-05-30T08:53:34Z", "topics": ["Multimodal"], "venueAttempts": [{"decisionRaw": "Accepted by The International Conference on Machine Learning (ICML 2026)", "evidence": [{"level": "author-claim", "observedAt": "2026-10-14T01:46:58.058199Z", "rawValue": "Accepted by The International Conference on Machine Learning (ICML 2026)", "sourceType": "arxiv-comments", "sourceUrl": "https://arxiv.org/abs/2606.00622"}], "reviewStatus": "accepted", "venueName": "International Conference on Machine Learning (ICML 2026)"}], "whyItMatters": "Addresses the lack of benchmarks for error propagation in long-horizon interactions, and shows existing mitigation methods are ineffective, motivating new approaches."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
