046211 - Deep Learning
General Information
The course is given in the Electrical and Computer Engineering Faculty at the Technion (ECE 046211) and covers several topics in deep learning (DL), with emphasis on supervised approaches to DL. Please visit our GitHub for the specific agenda and topics.
Winter 2024 Student Projects
| Project Title | Students | TL;DR | GitHub/YouTube |
|---|---|---|---|
| Model Compression and LoRA | Ido Blayberg & Ohad Amsalem | We tested the effectiveness of adding LoRA layer to a pre-trained compressed model for image classification task | GitHub Link YouTube Link |
| Suicidal Prediction Transformer | Uri Koren & Daniel Ochana | Predicting the need for suicide watch based on text, using transfomers (from scratch, GPT2, ChatGPT) | GitHub Link YouTube Link |
| Facial Emotion Detection Using GAN-generated Emotion Augmentations | Reuven Smbatyan & Amit Shalev | Leveraging EmotionGAN to enhance training process by creating emotion augmentations and expanding/balancing our dataset | GitHub Link |
| Song Sentiment Analysis | Zohar Milman & Dor Aviv | Fine-tuning of BERT model for the purposes of song lyrics sentiment analysis using LoRA and hyper-parameter sweep | GitHub Link YouTube Link |
| Predicting A Song’s Popularity Using Mamba | Itamar Horowitz & Rebecca Azoulay | Implementing a Mamba architecture in order to predict the popularity of a song | GitHub Link YouTube Link |
| DSP in Deep Learning – Morse Code Detector + Decoder | Maor Assayag & Eliraz Kadosh & Eliram Amrusi | We show that LSTM-RNN decodes Morse code at -5dB SNR with 2% CER, and Faster-RCNN (RESNET) achieves 98% IoU-x accuracy for Morse code detection | GitHub Link YouTube Link |
| Earthquake Prediction | Tzvi Tal Noy & Daniel Levi | Forecasting seismic activity, aiding in disaster management and mitigation efforts using RWKV | GitHub Link |
| Age Estimation from Chest X-ray | Iggy Segev Gal & Eyal Gilron | Experimenting with different transfer learning methods to estimate the age of a person from his chest Xray image | GitHub Link |
| Sherleaf-Holmes - Plant Species and Disease Type Classifier | Lee Benyamin & Ben Segal | Exploring different typs of architectures using transfer learning on image classification task - VGG, Alexnet, Resnet, Densenet, ViT + Lora & Dora | GitHub Link YouTube Link |
| GPT2-based sEMG Gesture Classifier | Yuval Gerzon & Stav Belyy | Transfer learning to classify EMG movement, overcoming small amount of labeled data, by training a GPT2 transformer-decoder model | GitHub Link |
| Music Genre Classification with Transformers | Yuval Hoffman & Roee Hadar | Using Transformers to classify music generes | GitHub Link YouTube Link |
| ViT vs. CNN for Elephants Classification | Ariel Lulinsky & Hadar Hai | Evaluating Convolutional Neural Networks (CNN) and Vision Transformers (ViT) for distinguishing between Asian elephants and African elephants | GitHub Link YouTube Link |
| Learning to Play 20 Questions | Yotam Norman & Shir Rotman | Leveraging pre-existing LLM capabilities for training a chatbot to play 20 questions | GitHub Link |
| This Image Does Not Exist… Or Is It? | Almog Anschel & Ran Elbaz | Using ViTs and Transfer Learning to evaluate if an image is generate by AI or not | GitHub Link |
| Basketball Action Recognition | Tal Dugma & Yonatan Ashlag & Yarin Bekor | Real-time recognition of basketball players and their actions on court | GitHub Link YouTube Link |
| Chaotic Time-Series Prediction using RNNs | Shaked Leslau & Rishona Daniels | Lorenz ‘63 chaotic time-series prediction using LSTM, GRU, Tranformers, and Reservor computing | GitHub Link YouTube Link |
Spring 2023 Student Projects
| Project Title | Students | TL;DR | GitHub/YouTube |
|---|---|---|---|
| Cartoonify- Cartonn Augmentation | Stav Lotan & Harel Mendelman | Testing the efficacy of “Cartoon Augmentations” for image classification | GitHub Link YouTube Link |
| MemeSense | Tomer Keniagin & Aviv Shem-Tov | Identifying hateful content in memes using combined text and image analysis | GitHub Link |
| Music Genre Classifier | Amit Karp & Ilay Yavlovich | Classifying songs into their respective genres using advanced models | GitHub Link YouTube Link |
| MRI Brain Tumor Patch-Based Classification | Nitzan Alt & Or Ronai | Detecting brain tumor types using MRI scans and transfer learning | GitHub Link YouTube Link |
| How I Feel | Itamar Nierenberg & Nir Elfasi | Recognizing human emotions from facial expressions in images | GitHub Link YouTube Link |
| RWKV-Based Music Generator | Ariel Suller & Liad Perl | Modifying LSTM-based music generation with an RWKV model | GitHub Link YouTube Link |
| Uncertainty Score for Medical Segmentation Models | Daniel Katz & Natalie Mendelson | Evaluating segmentation model performance with uncertainty scores | GitHub Link YouTube Link |
| ASL Translation from Images to Letters | Hadar Shloosh & Hadas Manor | Translating American Sign Language (ASL) images into English letters | GitHub Link |
| FeelMyText : Emotion Classifier From Text Using Data Augmentations | Yossi Meshulam & Eran Yermiyahu | Classifying emotions within raw textual content with augmented data | GitHub Link YouTube Link |
| Stock Price Prediction Using RWKV | Roee Latzres & Tomer Krichli | Forecasting stock prices using RWKV and numerical prediction | GitHub Link |
| AI-based Position Estimation | Avichay Ashur & Itay Geva | Predicting device location based on radio frequency data | GitHub Link YouTube Link (HEB) YouTube Link (ENG) |
| Facial Attribute Classification for Personalized Emoji Generation | Eitan Gorbonos & Amiel Gorbonos | Creating personalized emojis based on facial attributes | GitHub Link |
| Music Genre Classification using Wav2Vec2 | Itai Allouche & Adam Katav | Improving music genre classification with sound data representation | GitHub Link |
| ASLetter – Generating ASL Images From Letters | Nir Ben Haim & David Levit | Generating ASL images from English letters | GitHub Link |
| Climbing Holds Classification | Ori Levi & Or Pomeranz | Accurate classification and segmentation of climbing holds | GitHub Link YouTube Link |
| Enhancing Dense Crowd Counting CNNs | Rebecca Azoulay & Chris Shakkour | Improving the accuracy of crowd counting CNNs | GitHub Link YouTube Link |
| Super Mario Play | Amir Mishael & Lavi Doron | Learn Super Mario Bros gameplay and predict next key press | GitHub Link |
| Self-Learner - STL10 classification using BYOL representation | Tamar Sde Chen & Hadar Rosenber | Demonstrating an 11% improvement in supervised learning classification accuracy by leveraging self-supervised learning | GitHub Link YouTube Link |
| Image2Recipe: Deep Learning Network for Image-to-Recipe Translation | Sagi Eyal & Loren Tzveniashvily | Outputting recipes from food images using Deep Learning | GitHub Link |
| Efficient Use of Visual Transformers | Matan Millionschik& Yael Zak | Enhancing visual transformers with quantization and early exit | GitHub Link |
| Prism Diffusion | Ron Raphaeli & Sofiia Kurbatova | Controlling colors in diffusion-generated images using conditional LoRA | GitHub Link |
| FSLC - Fingerspelling Sign Language Classification | Samer Khair & Alam Shomary | Classifying fingerspelling sign language using CNNs | GitHub Link YouTube Link |
| Human Activity Recognition using Accelerometers data | Yair Stolero & Shlomi Buhadana | Comparing models for Human Activity Recognition from accelerometer data | GitHub Link |
| Style is All You Need | Evgeniy Pukhov & Dolev Hagai | Image style transfer using CNNs and optimization techniques | GitHub Link YouTube Link |
| BrainSight: Leveraging Vision Transformers for Tumor Detection | Gil Litvin, Ari Shemesh & Omer Paz | Achieving 99.61% accuracy in brain tumor classification using Vision Transformers | GitHub Link |
| Book Review Score Predictor | David Cojocaru & Ori Hagler | Predicting book review scores using NLP | GitHub Link YouTube Link |
| Image Captioning with LSTM&GRU | Meir Lederman & Shahar Alpert | Captioning images using Encoder-Decoder with LSTM and GRU | GitHub Link |
| Human or GPT: 2023 Turing Test | Noam Kasten & Muhammed Abu el Hija | Classifying text as human-authored or generated by GPT | GitHub Link YouTube Link |
Winter 2023 Student Projects
| Project Title | Students | TL;DR | GitHub/YouTube |
|---|---|---|---|
| ViT-BYOL | Shahar Yadin & Noa Shoham | Easily training a classifier with robustness to noise, without the need for explicitly showing it noisy images during training. | GitHub Link YouTube Link |
| Adversarial Attacks on Face Recognition Models | Shelly Francis & Gil Kapel | Defencing Adversarial Attacks on Face Recognition Models. | GitHub Link YouTube Link |
| Audio-based Instrument Detection | Roy Steinberg & Rotem Elimelech | With careful pre-processing and choice of loss functions, we predict multiple instruments in a musical piece. | GitHub Link YouTube Link |
| TransferMPL with Roleplaying | Shira Lifshitz & Saar Stern | We propose a “RolePlaying” mechanism and improve the “Meta pseudo Labels” method. | GitHub Link YouTube Link |
| Diffusion-based Data Augmentations | Eran Avneri & Itay Lamprecht | We show that using generative data augmentations with diffusion models improve the performence of CNN models. | GitHub Link YouTube Link |
| Black and White Image Colorization | Aviv Ish Shalom & Salomon Malka | We use a UNet architecture and test various loss functions for colorizing grayscale images. | GitHub Link YouTube Link |
| CLIP Your Food | Elizabet Khaimov & Ori Zehngut | We extract list of ingredients from a given input dish image using CLIP features. | GitHub Link YouTube Link |
| Limited-data Face Classification using Transfer Learning | Amit Kadosh & Batel Shuminov | Using a limited collection of face images we improve face recognition with transfer learning. | GitHub Link YouTube Link |
| Weights Pruning of Resnet50 Model | Michael Berko & Naomi Shapiro | We’ve managed to reducre nearly 40% of model’s parameters without changing its accuracy. | GitHub Link YouTube Link |
| Contrastive-center Loss for Speaker Recognition | Lior Bashari & Yonatan Kleerekoper | Examining the effect of Contrastive-center loss regularization on Speaker Recognition using transfer learning with ResNet-18. | GitHub Link YouTube Link |
| Football Player Detection | Avishav Engle & Daniel Hassid | Fine-tuned YOLOv7 NN to localize and classify footballer players and trained a BYOL network to cluster the teams. | GitHub Link |
| RNN and Transformer-based Image Captioning | Itai Shufaro & Nir Luria | We implemented an encoder-decoder network for image captioning. The encoder is a pre-trained CNN, and for the decoder we used both LSTM and Transformer networks. | GitHub Link YouTube Link |
| Generating Captions for Visual Stimuli from fMRI Scans | Yoav Tsoran & Roey Shafran | We use a pretained fMRI encoder and GPT2 to generate a caption of the image presented to a subject during an fMRI scan. | GitHub Link> YouTube Link |
| IMDB Score Prediction from a Movie’s Trailer | Eran Mann & Zeev Zukerman | Predicting IMDB score from selected frames from the movie’s trailer. | GitHub Link YouTube Link |
| ECG Classification with CNNs | Sagie Badoach & Gilad Altshuler | We classify ECG heartbeat to different disease classes, with simpler yet accurate (+98%) CNN model, and adapted the model to 2 tasks. | GitHub Link YouTube Link |
| Image Caption Generator | Dana Rip | We compared two architectures for image captioning: LSTM and Transformer. | GitHub Link YouTube Link |
| Boosting Classification with Estimated Depth | Mousa Arraf & Chen Katz | We boosted the results of classification models by using estimated depth as additional features. | GitHub Link |
| Augmenting Textual Datasets with GPT | Almog Zur & Ari Granevich | We used GPT2 to augment the IMDB dataset. | GitHub Link |
Winter 2022 Student Projects
| Project Title | Students | TL;DR | GitHub/YouTube |
|---|---|---|---|
| Adaptive STFT | Noam Elata & Rotem Idelson | Expanding on existing application of image processing networks to audio using STFT, we propose an adaptive STFT layer that learns the best DFT kernel and window for the application. | GitHub Link YouTube Link |
| Meta-Learning with Non-uniform Task Weights | Lior Friedman & Yair Nahum | We examine the effects of various task orderings and selections on meta-learning accuracy for images. | GitHub Link YouTube Link |
| Weapon Detection Using Transfer-Learning | Alon Nemirovsky & Itamar Ginsberg | We use YOLOv5 pre-trained model to detect threats in the form of pistols/knives using a dataset of life-like scene images. | GitHub Link YouTube Link |
| Sentence Transformer-VAE | Nofit Segal & Dan Haramati | We built a sentence VAE using the Transformer encoder-decoder architecture presented in “Attention Is All You Need” by Vaswani, Ashish, et al. | GitHub Link |
| EEG Classification | Nitzan Shitrit & Priel Salomon | We implement a CNN for classification of EEG signals recorded while a subject is viewing image of digits from 0 to 9 as stimuli. | GitHub Link YouTube Link |
| Beyond Inpainting - Combining Semantic Segmentation with Inpainting | George Pisha & Yevgeniy Men | We’ve implemented an end2end image segmentation and inpainting pipeline for auto-editing of images and extend the framework to video automatic object removal. | GitHub Link |
| Bitcoin Price Prediction Using Transformer | Yuval Baruch & Yuval Aidan | Predict Bitcoin short term price based on it’s former values using transformers architecture. | GitHub Link YouTube Link |
| Bird Species Classification & Noise Robustness | Hodaya Rabinovich & Yael Ilan | We focused on several pre-trained classification architectures and tried to ‘attack’ them with noise, comparing the results. Finally, we tried to improve the result using augmentations. | GitHub Link |
| Object Detection of Handwritten Circuit Diagrams | Dan Ilan Ben David & Adam Soker | Detecting and converting handwritten circuit diagrams to computer circuit diagram (LTspice). | GitHub Link YouTube Link |
| Facial Expression Recognition with Attention Prior | Ido Terner & Yair Gat | We have explored using an attention prior for facial expression recongnition using an attention prior. | GitHub Link |
| Malicious URL Detection | Edan Kinderman & Lev Panov | We detect Malicious URLs with LSTM and transformer architectures using character-level embedding methods. | GitHub Link |
| Two (Gyros) is All You Need | Gal Ness & Elad Zohar | We use the XY timeseries of smartwear accelometers to predict the Z motion with Transformer and novel positional encodings. | GitHub Link |
| Lips Don’t Lie | Tom Bekor & Mitchell Butovsky | Lip reading of a talking person without the use of audio, using facial-landmarks recognition and Transformer. | GitHub Link 1 GitHub Link 2 |
| Face Aging with StyleGAN2-ADA | Moshe Rafaeli & Udi Gal | We have implemented a conditional image generation model to perform face aging with a StyleGAN2-ADA generator and n InceptionResnetV1 encoder. | GitHub Link |
Spring 2021 Student Projects
| Project Title | Students | TL;DR | GitHub/YouTube |
|---|---|---|---|
| Cassava Leaf Deasses Classification | Ido Gabay & Aviya Cohen | Classification of imbalanced dataset from Kaggle with several methods, using transfer learning and our own conv-net. | GitHub Link YouTube Link |
| Traffic Signs Classification | Kfir Levi & Li-or Bar David | Comparing performances of different architectures on a traffic signs dataset. | GitHub Link YouTube Link |
| Short Term Stock Price Prediction with a Simple Trading Bot | Orel Tsioni & Roee Ben Shlomo | Predicting a stock price 1 minute into the future and trying to achieve a profit using a bot. | GitHub Link |
| Underwater Gesture Classification | Or De Goede & Matan Topel | Classifying gestures of divers in changing water conditions with 97.85% accuracy on the testset, using YOLOv5 for localization and our own CNN for classification. | GitHub Link |
| Spoken Digit Generator | Aviv Eliav & Ran Ben-Shaul | Generating audio files of spoken digits, using a conditional generative architectures (cVAE, cGAN) and evaluating the results with Inception Score. | GitHub Link |
| Using GAN generated images to improve classification small datasets | Daniel Bracha & Vladimir Kulikov | Using StyleGAN2-ADA to artificially generate more training images from the base dataset and comparing classification accuracy with and without the artificial data. | GitHub Link |
| RMR prediction Using Body Composition and Metabolic Adaptation Features | Amir Kfir & Tomer Keren | Based on private-use medical dataset, we build a DL architecture to predict Metabolic Rate values on rest based on other body-features. We also compare our results with classic ML algorithms. | GitHub Link |
| Sketch Classification | Roie Reshef & Gefen Dawidowicz | Sketch Classification on the “Quick, Draw!” dataset. Improves accuracy of an existing network by adding a special noise to the sketches called “shake pen”. | GitHub Link |
| Triplet Network for Few Shot Learning | Daniel Teitelman & Gonen Weiss | Implementing the triplet network for metric learning than using the learned feature vectors for few shot learning on the FashionMNIST dataset. | GitHub Link YouTube Link |
| Playing Chrome’s Dino Run with Deep Reinforcement Learning | Ari Frummer & Gal Kinberg | Comparing the performence of several DQN architectures and methods on the Dino Run game, including a classical DQN, a Dueling DQN, and using data augmentations. | GitHub Link |
| COVID19 Detection from CT Scans | Lin Sinorodin | Compare the performance of CNN vs. Vision Transformer as feature extraction models on image classification task (using transfer learning). | GitHub Link |
| Predicting the NASDAQ 100 using LSTM and GRU network | Roy Mahlab & Tamir Haver | Predicting the NASDAQ 100 index 1 week ahead using LSTM/GRU on the Federal Reserve total assets compared to a naive classifier. | GitHub Link |
| Image Caption with Attention | Neria Uzan & Razi Facheldeen | Using CNN and RNN methods build a network that can describe an image content. | GitHub Link |
| Music Genre Classifier | Omer Cohen & Jonathan Nir Shalit | Music genre classification using several Deep Learning architectures and methods. | GitHub Link YouTube Link |
| Mutual Information In Stock Prediction | Aviv Ratzon & Chay Guez | Training Transformer to predict next day closing prices of various stocks, and then adding information about related stocks and measure change in performance. | GitHub Link YouTube Link |
| Image Captioning | Luiz Wainstein & Yahel Kleinman | Using Encoder Decoder networks to generate descriptive sentences on image input. | GitHub Link |
| Sentiment Analysis of Machine Summarized Text | Roy Elkabetz | Does google’s pre-trained T5 general NLP model preserve sentiment of text when used as a summarizer? Let’s find out. | GitHub Link |
| Covid 19 Prediction from Chest X-Ray Images | Ariel Weizman & Yael Shavit | Using Transfer Learning from a model we pre-trained on detecting Pneumonia to the task of identifying COVID-19. | GitHub Link |
| UnmaskMe - Unmasked Face Generation | Ron Vainshtein & Shay Shimonov | Generating unmasked face images from masked face images using pix2pix cGAN architecture. | GitHub Link YouTube Link |
Contact
For more information/questions, pleaes contact Tal Daniel.