I believe it is because the tests conclusions that you have found being drawn by the practitioners are not without fault. There are two possible solutions. Fix the testing and or fix the causes in the learning systems. Obviously, any on who is making money stays out of either territory by necessity. You are making super contributions to places where you post. One of the most enjoyable processes is narrowing down the area of concern where work must be done to achieve solutions. It is important to continually change the nature of problems by whittling away at the boundaries. These days we are seeing the quant era coming to a close and it is particularly expensive as well. As we step past the sensory limitations of humans by introducing tools that extend human perception a whole new arena of pool extraction appears. What seems to be most pertinent is the determination of the end effects of segments of extraction. Leverage is fun as we all know but a greater opportunity lies with the leverage of extraction by dealing with how extraction can be a huge multiple of the "apparent" capacities of markets. A leverage of five doe lie within human perception quite easily. After that it get to be even more fun. It is never very difficult to manually analyze the data feeds since all data is from a finite series and secondly all data is binary. With the power od binary data, it is never necessary to use probability in any application whatsoever. This factor alone is sufficient to bring the quant era to a close. The ultimate goal is to simple extract a multiple of the "apparent" capacity of the market's offings. The data is binary so, therefore , the extraction is likewise. why would anyone go to any length whatsoever to do conversions of data and do conversiions of convesions back to binary executions? First of all it may be time consuming. you can see that finally, CERN is going to deal with things at hand by simply dumping unnecessary data instead of processing it. admittedly they have no time pressure and can simply store only pertinent data for later examination. Pool extraction it diffent simply because precision is important with respect to taking all that is available (or just the portion that is desired) at a given time. The quants have shown us, amply, how to freeze markets to the final illiquidity. The strategies that are replacing quant strategies are one's that inherently increase liquidity as a primary functionality. The longest it has ever taken me to find screw ups in the applications of actuary stuff is about three months. The shortest period of time is has taken the creators to put in "fixes" is more than twice as long. they have to go through a period, first, of determining how to translate the test results I have provided. This is because I do testing in manners that the SOP finds to be foreign. In the advancement of any field, there is not continuity. It is a case of the two differnt human enterprises coming into play: gathering and hunting. Neither has the characterisitics of the other in the same natural setting. So it is with testing. The manner of extraction, done mechanically, has no learning front end nor really any continuity of refinement. Refinement, largely, comes from redefining the problem and then summoning up the power to conduct the extraction. It, in fact, is a series of plateaux. AI, wholly or in its component artifices, does not in any way resemble rising plateau to plateau. I for one, get to look at the situation from the opposite side of the situation than most. I, from their vantage point, am in lala land except for one thing. I do what I am talking about and it comes to the bottom line the way I suggest that it does. There is a major fork in the road in the financial industry. One accumultes capital from a client base and bigger and bigger until there is no stastically significant difference between real wealth and the perfomance of the collection of client's capital. In effect, at best the client's capital floats along at the collective level of the markets and the related index. Proof of this is a pragmatic continuing observation. Obviously bouses result form this and the business of moving capital from one place to another to create and destroy various business operations. The other fork is removing capital from these pools as the markets afford that opportunity. AI is not currently participating in enhancing this strategy. It is not a case of detecting imbalance in the markets and profiting from imbalances. Quite the opposite, it is the case of extracting capital form the flow of capital. Obviously, the quants have recently and broadly proven that by inventing securitizations that are unidirectional and, thus, the application of capital brings the flow to zilch and illiquidity. Artificial adjustments to parallel standards to this securitization are not going to create liquidity because of the commensurate "corporate books'" synchronized strategies that are synchronized with the securitization. The capital is simply parked and periodical evaporated by the market to market (and sometime mark to model) processes. And new clients are invoked to offset the evaporation to prevent freezing the operators whole set of sectors. Here AI will tell anyone that the illiquidity is a permanent fixture of the securitization and corporate book strategy. The second fork, pool extraction, is foriegn to the business and to AI folks, so far. Pool extraction does not deal with market imbalances, it deals with opportunity to extrace the flow by dealing, exclusively with the flow. There is some "What wasn't that?" involved. NN deals largely with "what was that?" on the other hand. AI largely "gathers". Pool extraction deals with plateaux that mosre so resemble hunting. Differnt tools are used for differnt unconnected purposes in pool extraction plateau by plateau. Think of the recognition of all the individual identifiable strategies going on. All the game available to the hunter. Not the berries in fields all lined up and rooted in their respective places to be gathered. AI sees cause and effect of repeated processes. Pool extraction stalks in the present only and conducts an ever varying approach to maintain advantage and periodically kill. Look at used books. See which titles cost more than new did. The hunting books do. Gathers books are the commonly mentioned ones on "popular" lists. AI does not go to where the wild game is. The wild game goes to where it feels safe and has resourses (look at client brochures). It is very apparent where the herds are feeding as time passes and seasons change. It is dynamic and unrooted as are crops and berry bushes. Try hitting a traget moving towards you compared to a target moving away. A target can do anything it wants and in any order of actions. It is best to fence in the limits and extract each opportunity. It is very different than averaging (gaining marcro knowledge) distances away from where the berry bushes are standing to find out how far berries can roll. Record books are built with the contents of big game hunters and not berry pickers who are very smart about picking stuff. Doing a test as to whether a hunter is born or made is a nothing test, especially if the test is considered to be meaningful. How many trading systems have to be used to find out is a hunter is made or born? The answer is more than one. Twenty?. A hundred? lol... How would one find out if humans or AI was superior? A mechanical hunting system is not going to use AI in its construction. Pool extraction is not the approach pool builders are going to use to profit their clients. Pool builders protect pools. Can you imagine what it is like to build pipelines out of pools? The interpool plumbing is where the flow is. Quants look for leaks; I imagine they will fix a lot of leaks. This is simply humorous to me. It does take a lot of detection stuff and tooling to find leaks and to fix them. It is two part: keep internal leaks fixed and find external leaks to redirect back to the home pool. The market is where all the valves are for directing flow. You do not need a pool to play. You only need a valve; preferably a mechanical valve that is designed to deal with the flow from any pool to any others. Anyone is free to take any flow that is available. Valves opporating in parallel are a hoot!!!! I concluded to keep my valves open all the time to get flow no matter which way it is flowing; I put them all along the pipes as well.