Unverified50% confidenceFactExact time
The FM-Recon framework uses a frozen foundation model image encoder to extract multi-scale features, which after adjustment by a trainable refinement network are injected into each layer of the unrolled reconstruction network.
1
Sources
50%
Confidence
Long-term
Relevance
11/17/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedRECON's core idea is to feed the tokens output by a pretrained visual encoder as conditioning signals directly into the later layers of the diffusion model during training.70% similarUnverifiedUnrolled reconstruction networks unroll iterative optimization algorithms into a finite number of neural network layers that alternately perform data consistency and regularization, where regularization is replaced by learnable modules and data consistency retains its analytical form.68% similarUnverifiedUsing the jointly predictive model as a head on top of a frozen CausalPFN is a modular repair strategy that preserves the original unit-level point estimation capability and only replaces the way intervals are constructed.66% similarUnverifiedThe most effective path to recovering skill documents from a LoRA adapter and a frozen base model is to have the model first perform the task and then reconstruct the text from behavioral differences, rather than decoding directly from the weights.66% similarUnverified如果 ReplaySSM 的方法能够落地,状态存储的优化意味着同样的显存可以支撑更长的上下文或更大的批处理规模,且可直接集成到现有推理栈中,无需重新训练模型。66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1179258API
curl https://kongchang.com/api/v1/knowledge/claims/1179258MCP
get_claim(id=1179258)