[
  {
    "text": "The Long Short-Term Memory",
    "score": 0.9994322657585144,
    "polygon": [
      [
        32.0,
        24.0
      ],
      [
        543.0,
        30.0
      ],
      [
        543.0,
        65.0
      ],
      [
        32.0,
        60.0
      ]
    ]
  },
  {
    "text": " LSTM networks learns long-term dependencies better",
    "score": 0.9935519099235535,
    "polygon": [
      [
        120.0,
        268.0
      ],
      [
        823.0,
        269.0
      ],
      [
        823.0,
        294.0
      ],
      [
        120.0,
        292.0
      ]
    ]
  },
  {
    "text": " Optimization",
    "score": 0.9938980340957642,
    "polygon": [
      [
        119.0,
        316.0
      ],
      [
        311.0,
        316.0
      ],
      [
        311.0,
        343.0
      ],
      [
        119.0,
        343.0
      ]
    ]
  },
  {
    "text": ": Clipping gradient",
    "score": 0.9587327837944031,
    "polygon": [
      [
        180.0,
        364.0
      ],
      [
        406.0,
        364.0
      ],
      [
        406.0,
        385.0
      ],
      [
        180.0,
        385.0
      ]
    ]
  },
  {
    "text": ". Regularizing: encourage information flow",
    "score": 0.9854925870895386,
    "polygon": [
      [
        177.0,
        399.0
      ],
      [
        685.0,
        400.0
      ],
      [
        685.0,
        428.0
      ],
      [
        177.0,
        427.0
      ]
    ]
  },
  {
    "text": ". Case studies:",
    "score": 0.8947544693946838,
    "polygon": [
      [
        119.0,
        446.0
      ],
      [
        311.0,
        448.0
      ],
      [
        311.0,
        472.0
      ],
      [
        119.0,
        470.0
      ]
    ]
  },
  {
    "text": " Memory networks (Westion et al., 2014)",
    "score": 0.9780460000038147,
    "polygon": [
      [
        181.0,
        494.0
      ],
      [
        681.0,
        494.0
      ],
      [
        681.0,
        518.0
      ],
      [
        181.0,
        518.0
      ]
    ]
  },
  {
    "text": " Neural Turing machine (Graves et al., 2014)",
    "score": 0.9989622235298157,
    "polygon": [
      [
        181.0,
        532.0
      ],
      [
        727.0,
        532.0
      ],
      [
        727.0,
        557.0
      ],
      [
        181.0,
        557.0
      ]
    ]
  },
  {
    "text": " Multiple object recognition with attention (Ba et al.)",
    "score": 0.9943037629127502,
    "polygon": [
      [
        182.0,
        569.0
      ],
      [
        830.0,
        571.0
      ],
      [
        830.0,
        595.0
      ],
      [
        182.0,
        593.0
      ]
    ]
  },
  {
    "text": ": Image captioning",
    "score": 0.97649085521698,
    "polygon": [
      [
        181.0,
        604.0
      ],
      [
        409.0,
        610.0
      ],
      [
        408.0,
        638.0
      ],
      [
        180.0,
        633.0
      ]
    ]
  },
  {
    "text": "ecturer: Duc Dung Nguyen, PhD. Contact: nddung@hcmut.edu.vn",
    "score": 0.9703457951545715,
    "polygon": [
      [
        50.0,
        761.0
      ],
      [
        506.0,
        761.0
      ],
      [
        506.0,
        775.0
      ],
      [
        50.0,
        775.0
      ]
    ]
  },
  {
    "text": "Deep Learning",
    "score": 0.996578574180603,
    "polygon": [
      [
        765.0,
        758.0
      ],
      [
        864.0,
        760.0
      ],
      [
        864.0,
        778.0
      ],
      [
        764.0,
        776.0
      ]
    ]
  },
  {
    "text": "47/ 47",
    "score": 0.9623189568519592,
    "polygon": [
      [
        1120.0,
        759.0
      ],
      [
        1175.0,
        759.0
      ],
      [
        1175.0,
        777.0
      ],
      [
        1120.0,
        777.0
      ]
    ]
  }
]