MO434 - Deep Learning

Computing Institute (IC / Unicamp)
Prof. Sandra Avila (sandra@ic.unicamp.br)

 

Time and Place: Mondays, from 14h to 16h. Classes will take place via Google Meet, with some classes being held synchronously and others asynchronously. For all synchronous classes, the videos of the classes will be made available later. Classes will start on September 21st / 2020.

Waiters The opening hours will be provided by the teacher and PED Tito Rezende in the attendance classes or by Slack (dl-unicamp-2020.slack.com).

Course Program: Introduction to Deep Learning, [Deep] Neural Networks, Convolutional Neural Networks (CNNs), Architectures, Training, Recurring Neural Networks (RNNs), Transformers, Detection and Segmentation, Adevrary Generative Networks, Visualization Techniques and Model Interpretation , FATE (Fairness, Accountability, Transparency, and Ethics) in AI.

Programming language: The programming language used in the course is Python. Tools: Keras & TensorFlow, PyTorch.

Assessment: The evaluation will be based on the proposed activities and practical projects, being:

• At the end of the class, the student must send via Google Classroom what was the point of biggest doubt of the class (D), a succinct question. If you have no doubts (really? Really?), The student should highlight the point that she found most interesting. Answers like “I had no doubts” or blank answers will not be accepted.
• Three double tasks, T1, T2 and T3. The code and the report must be delivered via Google Classroom.
• Um final project PF to be carried out in groups:
  • The (O) student (o) will only be able to do the final PF project if she has delivered the three tasks T1, T2 and T3.
  • The groups must have 3 students (the), necessarily.
  • The code and the report must be delivered via Google Classroom, and the report must present an explanation of the technique implemented, illustrations of the results, and a discussion of the results obtained in the format of a scientific article, in the model suggested by the teacher.
  • The project must be presented (in a 4-minute video format), by the group, on the scheduled date.
• The final average, M, will be calculated as: M = 0,15 x T1 + 0,15 x T2 + 0,15 x T3 + 0,5 x PF + 0,05 x D
  • For undergraduate students, the following rule will apply:
    • Approved: if M ≥ 5.0
    • Failed: if M <5.0
  • For graduate students, the final concept will be attributed as follows:
    • A: if M ≥ 8.5
    • B: if 7.0 ≤ M <8.5
    • C: if 5.0 ≤ M <7.0
    • D: if M <5.0

Submission of Activities: All course activities must be submitted via Google Classroom.

Evaluation Delivery Dates: The dates below are subject to change.

• Biggest Doubt (D): Until 12:XNUMX, three days after class.
• Task 1 (T1): 26/10/2020
• Task 2 (T2): 23/11/2020
• Task 3 (T3): 21/12/2020
• Final Project (PF):
  • Proposal submission (theme and database) (10% PF): 12/10/2020
  • Baseline submission (first result) (20% FP): 07/12/2020
  • Presentation (videos up to 4 minutes) (30% PF): 11-18 / 01/2021 Examples 2019.2
  • PF submission (report and code) (40% PF): 11/01/2021

Remarks:

• There will be no tests or exam for this discipline.
• Any attempted fraud in the activities of the discipline will imply a final average M = 0 (zero) for all persons involved, without prejudice to other sanctions.

References: The teacher will not follow a specific textbook, however, the following books cover what will be seen in class:

• "Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow", A. Géron, 2019.
• "Deep Learning", I. Goodfellow, Y. Bengio, & A. Courville, 2016.
• “Dive into Deep Learning”, M. Gardner, M. Drummy, J. Quinn, J. McEachen, & M. Fullan, 2019.
• "Deep Learning with PyTorch", E. Stevens, L. Antiquity, T. Viehmann, 2020.