**Temporal Attention-based Long-Short Term Memory (TALSTM) C

Temporal Attention-based Long-Short Term Memory (TALSTM) Challenge

In this challenge, we'll push the limits of Long Short-Term Memory (LSTM) networks by introducing a novel temporal attention mechanism. Your task is to design, train, and evaluate a TALSTM model that can learn to forecast the trajectory of a dynamic system with multiple, interacting components.

Constraints:

1. Multi-modal input: Your model will accept three types of input data:
   • Time-series data (e.g., sensor readings) with 10 features
   • Image data (e.g., camera feeds) with 3 color channels
   • Text data (e.g., system logs) with 5 keywords
2. Dynamic system: The system consists of 5 interacting components, each with its own temporal dynamics. The components are:
   • Component A: Temperature (time-series)
   • Component B: Humidity (time-series)
   • Component C: Air pressure (time-series)
   • Component D: Visual sensor (image)
   • Component E: System alert logs (text)
3. Forecasting horizon: Your model must predict the values of each component for the next 24 hours.
4. Model evaluation: We'll use a combination of metrics, including mean absolute error (MAE), mean squared error (MSE), and precision@k for the top-k most accurate predictions.
5. Temporal attention mechanism: Your model must incorporate a novel temporal attention mechanism that adapts to the varying temporal relationships between the interacting components.
6. Model size: The total number of model parameters must not exceed 100 million.

Dataset:

We provide a pre-processed dataset consisting of 1 million samples, each with the three input modalities and 5 target components. The dataset is available on a private cloud storage service.

Evaluation metrics:

• Mean Absolute Error (MAE)
• Mean Squared Error (MSE)
• Precision@k (k=5)

Submission guidelines:

1. Implement your TALSTM model using a deep learning framework (e.g., PyTorch, TensorFlow).
2. Train your model on the provided dataset for a maximum of 100 epochs.
3. Evaluate your model using the provided evaluation metrics.
4. Submit your model implementation, along with the evaluation results, to a designated GitHub repository.

Deadline:

Submissions are due on December 15, 2025. Late submissions will not be accepted.

Prizes:

The top 3 submissions will receive prizes of $10,000, $5,000, and $2,000, respectively.

Are you up for the challenge?

---

Publicado automáticamente