RESUMO
Deep cooperative multi-agent reinforcement learning has demonstrated its remarkable success over a wide spectrum of complex control tasks. However, recent advances in multi-agent learning mainly focus on value decomposition while leaving entity interactions still intertwined, which easily leads to over-fitting on noisy interactions between entities. In this work, we introduce a novel interactiOn Pattern disenTangling (OPT) method, to disentangle the entity interactions into interaction prototypes, each of which represents an underlying interaction pattern within a subgroup of the entities. OPT facilitates filtering the noisy interactions between irrelevant entities and thus significantly improves generalizability as well as interpretability. Specifically, OPT introduces a sparse disagreement mechanism to encourage sparsity and diversity among discovered interaction prototypes. Then the model selectively restructures these prototypes into a compact interaction pattern by an aggregator with learnable weights. To alleviate the training instability issue caused by partial observability, we propose to maximize the mutual information between the aggregation weights and the history behaviors of each agent. Experiments on single-task, multi-task and zero-shot benchmarks demonstrate that the proposed method yields results superior to the state-of-the-art counterparts. Our code is available at https://github.com/liushunyu/OPT.
RESUMO
OBJECTIVE: To evaluate the roles of folic acid and beta-carotene in the chemoprevention of gastric and other gastrointestinal (GI) cancers. METHODS: In a randomized, double-blind, placebo-controlled trial, a total of 216 patients with atrophic gastritis were randomly assigned to one of the four groups: (1) folate (FA, 20 mg per day plus vitamin B(12) 1 mg, intramuscularly, per month for one year, then 20 mg two times a week plus 1 mg per three months for the next year); (2) natural beta-carotene (N-betaC, 30 mg per day for first year, then 30 mg two times a week for the next); (3) synthetic beta-carotene (S-betaC, administered as in N-betaC); and (4) placebo. Follow-ups continued from 1994 to 2001. RESULTS: A total of 7 new cases of gastrointestinal cancers were diagnosed with 3 stomach, 1 colon and 1 esophageal cancers occurring in the placebo group; 1 stomach cancer in both of the N-betaC and S-betaC groups, and no cancer occurring in FA group. In terms of GI cancers, there was a significant reduction in the FA group, compared with the placebo group (P = 0.04). A similar trend was observed in both N-betaC and S-betaC groups (P = 0.07 - 0.08). Taken together, the three intervention groups displayed a highly significant decrease in occurrence (P = 0.004, vs placebo), and a lower risk for GI cancers (OR = 0.12; 95% confidence interval, 0.03 - 0.51). For development of gastric cancer, any one of the three active-treated groups did not reach statistically significant reduction. The FA group showed obvious improvement of the gastric mucosal lesions with more patients displaying lesions reversed or stable atrophy and inflammation (P = 0.04), reversed intestinal metaplasia (P = 0.06) at the end of follow-up, and reversed displasia (P = 0.017) at 12 months. Two cases of false jaundice were found in beta-carotene groups with no influence on administration, and no side-effects were reported in FA group. CONCLUSIONS: This trial revealed the interventional effect of folic acid on the development of GI cancers, a similar effect of beta-carotene was also detected. Also, folic acid may be of use to treat atrophic gastritis by preventing or reversing the precancerous lesions.