Downscaling GLDAS soil moisture data based on MODIS-Lai

I found few recent research paper that are downscaling very high resolution soil moisture/precipitation data based on MODIS-Lai using machine learning methods. However, I was not able to find the code anywhere and write the code myself in Earth Engine. But, I made a code which make 3-hourly data to daily data. What is the next step? How can I downscale GLDAS using machine learning methods?

   // Start and End Dates
    var inidate = ee.Date('2000-02-18');
    var enddate = ee.Date('2020-12-31');

    var modislai = ee.ImageCollection('MODIS/006/MCD15A3H')
        .select('Lai')
        .filterDate(inidate, enddate);  

    // Difference between start and end in days 
    var difdate = enddate.difference(inidate, 'day');

    //Import GLDAS data  
    var imgcol = ee.ImageCollection("NASA/GLDAS/V021/NOAH/G025/T3H")
            .select('SoilMoi10_40cm_inst')
            .filterDate(inidate, enddate);
    
    // Time lapse
    var lapse = ee.List.sequence(0, difdate.subtract(1))
    var listdates = lapse.map(function(day){
      return inidate.advance(day, 'day')
    })
    
    var daily_gldas = ee.ImageCollection.fromImages(listdates.map(function(day){
      day = ee.Date(day);
      var daily_collection = imgcol.select('SoilMoi10_40cm_inst')
                                    .filter(ee.Filter.date(day, day.advance(1, 'day')));
      return daily_collection.mean().rename('daily_sm_mean')
                              .set('system:index', day.format('YYYY-MM-dd'))
                              .set('date', day.format('YYYY-MM-dd'))
                              .set('system:time_start', day.millis())
                              .set('system:time_end', day.millis())
      }));