Education

Skills

• 
• 
• 
• 

• Bilingual Arabic - French
• English Fluent speaker, good in reading and writing (TOEFL: 103)

Research Interests & Publications

In general, my research focuses on studying deep neural networks behavior in small sample regimes. For this purpose, I investigate regularization methods incorporating different priors in the learning objective. From application point of view, I am mainly interested in medical image analysis where the small sample regime is usually an issue.

Learning from small data:

In this research axis, we aim to bridge the gap between statistical learning theory and deep neural network developments. While the former provides solid foundations for regularization of several learning algorithms such as SVMs and logistic regression, deep neural networks still suffer from a lack of systematic and mathematically motivated regularizers. In this context, we propose to use a regularizer inspired from classical learning theory: Function norm. In other words, we limit the hypothesis set in which we optimize the objective to a ball in the L₂ function space. As we proved that the exact computation of this norm is NP-hard, we propose to estimate it stochastically by generating samples using a variational autoencoder.

For more details:

Stochastic Function Norm Regularization of DNNs -
Amal Rannen Triki & Matthew B. Blaschko
NIPS workshop on optimization for machine learning - OPT2016

Stochastic Weighted Function Norm Regularization -
Amal Rannen Triki & Matthew B. Blaschko
Work in progress

Learning continually:

In this research axis, we consider the problem of lifelong learning. For example, when data from a new domain is presented to a neural network, the weights of the network adapt drastically to the new structure and tend to forget what was previously seen. We propose a solution where a single model is trained for a sequence of tasks. Our method aims at preserving the knowledge of the previous tasks while learning a new one by using autoencoders. For each task, an under-complete autoencoder is learned, capturing the features that are crucial for its achievement. When a new task is presented to the system, we use the autoencoders to regularize the training by preserving the information on which the previous tasks are mainly relying. At the same time, as the used autoencoders are under-complete the features are given space to adjust to the new data structure as only low dimensional submanifold is controlled.

For more details:

Encode Based Lifelong Learning -
Amal Rannen Triki^*, Rahaf Aljundi^*, Matthew B. Blaschko & Tinne Tuytelaar
ICCV 2017

Learning from medical data:

In this research axis, we apply the methods developed in the previous axes to real-life medical applications.

In order to test DNNs with function norm regularization, we considered the problem of assessing the surgical margin during breast lumpectomy operations, which can avoid the need for additional surgery. Optical coherence tomography (OCT) is an imaging technique that has been proven to be efficient for this purpose. However, to avoid overloading the surgeon during the operation, automatic cancer detection at the surface of the removed tissue is needed. Some methods based on the spatial statistics of the images have been developed, but the obtained results are still far from human performance. In our work, we investigate the possibility to use deep neural networks (DNNs) for real time margin assessment, demonstrating performance significantly better than the reported literature and close to the level of a human expert. Since the goal is to detect the presence of cancer, a patch-based classification method is proposed, as it is sufficient for detection, and requires training data that is easier and cheaper to collect than for other approaches such as segmentation. As the number of available images is small with comparison to the number of weights in the network, we use function norm regularization as introduced in Stochastic Function Norm Regularization of DNNs. We show a significant improvement over the best results in the literature, and a better behavior than Weight Decay and DropOut.

For more details:

Intraoperative margin assessment of human breast tissue in optical coherence tomography images using deep neural networks -
Amal Rannen Triki, Matthew B. Blaschko, Yoon Mo Jung, Seungri Song, Hyun Ju Han, Seung Il Kim & Chulmin Joo
Work in progress

Experience

Awards

• Tunisian National Excellence Scholarship – Among 20 first candidates nationally for the Baccalaureate exam
• Finalist for the Qualcomm Innovation Fellowship – Europe – 2016