This week's shameless go with the boss of the ChinaSys, nb can be said that most of the entire group of people of China to engage in academic exchange system. A lap down to listen to the report, there are a few ideas, not much, probably not so deep.
Open source framework system in the field is not much
to engage the system and engage in AI, engage in different algorithms, systems do not only need a good Idea, more attention can achieve it. Before contact with AI, often say that someone of the same open-source data train does not come out the same results, but is now engaged in the system, more it is not readily available framework to engage, if doing system-level experiments may be needed from start to realize their own source code, the code requires very high capacity. AI is not possible until now the results of the same complex, now is not even ran the experiment. May be specific to a particular field, the focus is also different between the groups, there would not be a set of reusable, to meet the needs of the majority of the overall framework occur.
The system also has a tendency to engage in AI and a combination of
several talk and AI are combined are down and listen to reinforcement learning combined use reinforcement learning to make certain predictions about the behavior of the subsequent year. As well as with reinforcement learning for DB do best config, the final effect is amazing good, even more than a deep experience of the DBA. But for DL this area, are now the focus of fire on GAN, I heard almost became iGAN the ICLR (233,333). At the same time feeling DL may make some improvement on some of the decisions on the system, then bring to further enhance system performance. But the system may do more or do trade off or do exploration in new areas, to enhance these aspects of AI may be not very strong. If you go to the direction at the same time that a violent train DL with traditional massive data sets, how to solve the problem of data sets also need to be considered. So if you are using AI, it may be more use of the RL method.
Do not miss doing research when small dots
next hear a teacher say the source of their paper turned out to be the group doing traditional research, we found a small technical point. So as to explore a new pit, quickly studied under, then out of the top one will. This article is about the ARM Barrier, the specific details are not really understand, but if I put it, may just avoid the pit usually works just fine on, I did not expect that they will continue to dig deep down, admirable. It may also require a certain sensitivity it hhhh
Optimization dare to look on the bottom
there is a systematic study of Block Chain papers, talking about how to find and optimize the performance of block chain system. Specific mathematical did not quite understand, but in the end turned out to be placed on the command level for analysis, and then get the size of the impact of different types of instructions, and find out about this part of the parallel instruction is not a major factor, up from the instruction set level to see such a large system, and then get an impressive conclusion is admirable.
In sum, the paper is done to share some greater degree of advance in the relevant fields, although the direction Gehangrugeshan, systematic study so much, but when I see other people's bright spot from the research of others to share in , is also a big future grow on their own level.
In addition the ability to really make the system code requirements of a good high ah 5555555