Unverified50% confidenceFactExact time
An object navigation policy trained in scenes generated by SceneBloom achieved a higher success rate on real scanned data not involved in training than using an equal amount of real training scans.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/26/2026
First Seen
Valid until: 1/24/2027
Sources
Related Entities
Related Claims
UnverifiedMethods such as freezing LLM backbone parameters during the visual alignment training phase, staged training, data ratio replay, and LoRA can mitigate catastrophic forgetting in multimodal training.70% similarUnverifiedLightweight detectors may need to be trained for specific models or specific tasks, and their cross-model transferability may be limited.70% similarUnverifiedBy projecting input activations onto a small subset of the principal components of the training distribution, probes can achieve cross-domain transfer with performance nearly matching that of probes trained directly on the test distribution.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1098359API
curl https://kongchang.com/api/v1/knowledge/claims/1098359MCP
get_claim(id=1098359)