Unverified50% confidenceTradeoffExact time
The computational cost of self-play training is higher than one-off fine-tuning, because each iteration must generate samples and train the discriminator.
1
Sources
50%
Confidence
Long-term
Relevance
11/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe DQN-paradigm reinforcement learning approach requires millions of steps of self-play training, consumes enormous resources, and the trained results are difficult to transfer to new games.75% similarUnverifiedTotal energy consumption on the inference side may exceed that of training, because it accumulates with user scale and invocation frequency.74% similarUnverifiedAdversarial training (AT) improves model robustness by generating adversarial examples during training and including them in training, but it incurs enormous memory overhead and involves a trade-off between clean accuracy and robust accuracy.74% similarUnverifiedThe experiments count encoder pre-training compute within the same lifecycle compute budget, so compute consumed by pre-training encroaches on the compute available for multimodal training.74% similarUnverifiedDuring training with mixed precision, memory usage roughly requires 16-20 bytes per parameter (including parameters, gradients, and optimizer states) plus activation overhead.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1173824API
curl https://kongchang.com/api/v1/knowledge/claims/1173824MCP
get_claim(id=1173824)