Gallery AI Exhibit Scorers Alpha

This is a small project to try to develop (using Genetic Algorithms) a set of simple Java classes for guessing an exhibit's likely appeal to Gallery users. In the first instance this is restricted to 2D still images.

Though this is based on classical GA techniques for finding optimal Scorers (eg random mutation, breeding, selection, etc) this also attempts to speed up the search for optimality by using conventional mathematical "error minimisation" techniques/algorithms (such as the FreeHEP Java Minuit minimiser/optimiser) widely used in numerical optimisation in the physical sciences and the financial markets for example. This brings the exotic skills of the numerical mathematicians to bear on the problem in tandem with the Artificial Intelligenciers!

One use of this is to help predict at or before upload of new images to the Gallery those that are likely to be most popular. The contributor may then choose to upload or not, and/or create more variations rather than waiting for bouquets or brickbats from the users.

Another possible use is to automatically produce a "tweaked" version of an existing exhibit more likely to appeal to Gallery visitors.

Status

The current set of available "best" exhibit Scorers (with a total population of 8046) is: [FixedScore:resultScore=72, LocalSampler:sampleSizePower=6:outlierFractionPower=5:component=Y:sampleCombination=RANGE:pixelSampleMethod=VALUE:S0=-992:S1=32678:S2=11011:S3=28531:S4=24358:S5=-18209:S6=7283:S7=32767:S8=17797:S9=-4795:Sa=25276:Sb=15266:Sc=11309:Sd=21213:Se=11304:Sf=-6569:Sg=19377:C0=32767:C1=32596:C2=32637:C3=32611:C4=32635:C5=32767:C6=32658:C7=32671:C8=32694:C9=32690:Ca=32523:Cb=32767:Cc=32767:Cd=32767:Ce=32727:Cf=32767:Cg=32750, LocalSampler:sampleSizePower=6:outlierFractionPower=5:component=Y:sampleCombination=RANGE:pixelSampleMethod=VALUE:S0=32767:S1=21328:S2=20112:S3=21955:S4=23443:S5=-6212:S6=852:S7=30439:S8=9794:S9=-4844:Sa=25706:Sb=15974:Sc=11515:Sd=19895:Se=11988:Sf=-6421:Sg=24115:C0=32767:C1=32767:C2=32767:C3=32767:C4=32756:C5=32713:C6=32458:C7=32762:C8=32757:C9=32660:Ca=32767:Cb=32767:Cc=32767:Cd=32767:Ce=32767:Cf=32767:Cg=32767, LocalSampler:sampleSizePower=6:outlierFractionPower=5:component=Y:sampleCombination=RANGE:pixelSampleMethod=VALUE:S0=32767:S1=32767:S2=11011:S3=21955:S4=23443:S5=-6212:S6=852:S7=30439:S8=17797:S9=-4844:Sa=25276:Sb=15974:Sc=11515:Sd=19895:Se=11988:Sf=-6421:Sg=24115:C0=32767:C1=32767:C2=32767:C3=32767:C4=32615:C5=32767:C6=32446:C7=32767:C8=32767:C9=32767:Ca=32767:Cb=32767:Cc=32767:Cd=32751:Ce=32767:Cf=32767:Cg=32767, RandomPositiveScore, SimpleExposure:sampleSizePower=3:brightnessRangePercent=47:outlierFractionPower=2, SimpleExposure:sampleSizePower=5:brightnessRangePercent=100:outlierFractionPower=9, SimpleExposure:sampleSizePower=9:brightnessRangePercent=93:outlierFractionPower=4, ZeroScore].

The current raw unweighted score of a popular JPEG (South Africa coins various values ZAR South African Rand R5 old and new bimetal R2 50c 10c worn tarnished back 1 DHD.jpg) by of each of the Scorers is as follows:

• FixedScore:resultScore=72: score=14417/32767:confidence=32767/32767
• LocalSampler:sampleSizePower=6:outlierFractionPower=5:component=Y:sampleCombination=RANGE:pixelSampleMethod=VALUE:S0=-992:S1=32678:S2=11011:S3=28531:S4=24358:S5=-18209:S6=7283:S7=32767:S8=17797:S9=-4795:Sa=25276:Sb=15266:Sc=11309:Sd=21213:Se=11304:Sf=-6569:Sg=19377:C0=32767:C1=32596:C2=32637:C3=32611:C4=32635:C5=32767:C6=32658:C7=32671:C8=32694:C9=32690:Ca=32523:Cb=32767:Cc=32767:Cd=32767:Ce=32727:Cf=32767:Cg=32750: score=12547/32767:confidence=32767/32767
• LocalSampler:sampleSizePower=6:outlierFractionPower=5:component=Y:sampleCombination=RANGE:pixelSampleMethod=VALUE:S0=32767:S1=21328:S2=20112:S3=21955:S4=23443:S5=-6212:S6=852:S7=30439:S8=9794:S9=-4844:Sa=25706:Sb=15974:Sc=11515:Sd=19895:Se=11988:Sf=-6421:Sg=24115:C0=32767:C1=32767:C2=32767:C3=32767:C4=32756:C5=32713:C6=32458:C7=32762:C8=32757:C9=32660:Ca=32767:Cb=32767:Cc=32767:Cd=32767:Ce=32767:Cf=32767:Cg=32767: score=12562/32767:confidence=32767/32767
• LocalSampler:sampleSizePower=6:outlierFractionPower=5:component=Y:sampleCombination=RANGE:pixelSampleMethod=VALUE:S0=32767:S1=32767:S2=11011:S3=21955:S4=23443:S5=-6212:S6=852:S7=30439:S8=17797:S9=-4844:Sa=25276:Sb=15974:Sc=11515:Sd=19895:Se=11988:Sf=-6421:Sg=24115:C0=32767:C1=32767:C2=32767:C3=32767:C4=32615:C5=32767:C6=32446:C7=32767:C8=32767:C9=32767:Ca=32767:Cb=32767:Cc=32767:Cd=32751:Ce=32767:Cf=32767:Cg=32767: score=12562/32767:confidence=32765/32767
• RandomPositiveScore: score=152/32767:confidence=152/32767
• SimpleExposure:sampleSizePower=3:brightnessRangePercent=47:outlierFractionPower=2: score=15370/32767:confidence=15370/32767
• SimpleExposure:sampleSizePower=5:brightnessRangePercent=100:outlierFractionPower=9: score=26316/32767:confidence=26316/32767
• SimpleExposure:sampleSizePower=9:brightnessRangePercent=93:outlierFractionPower=4: score=25800/32767:confidence=25800/32767
• ZeroScore: score=0:confidence=32767/32767

The current "ratings" of each of the best and base Scorers is as follows:

• FixedScore: score=815/32767:confidence=27030/32767 [0.020517776]
• FixedScore:resultScore=72: score=4254/32767:confidence=27030/32767 [0.107095234]
• LocalSampler: score=1411/32767:confidence=20016/32767 [0.026304554]
• LocalSampler:sampleSizePower=6:outlierFractionPower=5:component=Y:sampleCombination=RANGE:pixelSampleMethod=VALUE:S0=-992:S1=32678:S2=11011:S3=28531:S4=24358:S5=-18209:S6=7283:S7=32767:S8=17797:S9=-4795:Sa=25276:Sb=15266:Sc=11309:Sd=21213:Se=11304:Sf=-6569:Sg=19377:C0=32767:C1=32596:C2=32637:C3=32611:C4=32635:C5=32767:C6=32658:C7=32671:C8=32694:C9=32690:Ca=32523:Cb=32767:Cc=32767:Cd=32767:Ce=32727:Cf=32767:Cg=32750: score=5483/32767:confidence=26800/32767 [0.136861]
• LocalSampler:sampleSizePower=6:outlierFractionPower=5:component=Y:sampleCombination=RANGE:pixelSampleMethod=VALUE:S0=32767:S1=21328:S2=20112:S3=21955:S4=23443:S5=-6212:S6=852:S7=30439:S8=9794:S9=-4844:Sa=25706:Sb=15974:Sc=11515:Sd=19895:Se=11988:Sf=-6421:Sg=24115:C0=32767:C1=32767:C2=32767:C3=32767:C4=32756:C5=32713:C6=32458:C7=32762:C8=32757:C9=32660:Ca=32767:Cb=32767:Cc=32767:Cd=32767:Ce=32767:Cf=32767:Cg=32767: score=5558/32767:confidence=26828/32767 [0.138878]
• LocalSampler:sampleSizePower=6:outlierFractionPower=5:component=Y:sampleCombination=RANGE:pixelSampleMethod=VALUE:S0=32767:S1=32767:S2=11011:S3=21955:S4=23443:S5=-6212:S6=852:S7=30439:S8=17797:S9=-4844:Sa=25276:Sb=15974:Sc=11515:Sd=19895:Se=11988:Sf=-6421:Sg=24115:C0=32767:C1=32767:C2=32767:C3=32767:C4=32615:C5=32767:C6=32446:C7=32767:C8=32767:C9=32767:Ca=32767:Cb=32767:Cc=32767:Cd=32751:Ce=32767:Cf=32767:Cg=32767: score=5586/32767:confidence=26834/32767 [0.13960886]
• MaxScore: score=-1139/32767:confidence=27030/32767 [-0.028674537]
• MinScore: score=-22365/32767:confidence=27030/32767 [-0.56304306]
• NoConfidence: score=0:confidence=0 [0.0]
• RandomPositiveScore: score=1876/32767:confidence=13633/32767 [0.023820503]
• RandomScore: score=-6164/32767:confidence=27030/32767 [-0.15517986]
• SimpleExposure: score=-929/32767:confidence=25445/32767 [-0.022016322]
• SimpleExposure:sampleSizePower=3:brightnessRangePercent=47:outlierFractionPower=2: score=4470/32767:confidence=13233/32767 [0.055092502]
• SimpleExposure:sampleSizePower=5:brightnessRangePercent=100:outlierFractionPower=9: score=3171/32767:confidence=17127/32767 [0.050582953]
• SimpleExposure:sampleSizePower=9:brightnessRangePercent=93:outlierFractionPower=4: score=2440/32767:confidence=16724/32767 [0.03800639]
• ZeroScore: score=815/32767:confidence=27030/32767 [0.020517776]

You can interact with the AI system on the remote page.