1 | Offline Policy Evaluation with Adaptively Collected Data | Sunil Madhow |
2 | An Exploration of Facial Emotion Recognition Models and Building a User-Friendly Interface | Gautam Mundewadi, Chinmay Sonar |
3 | How should I pronounce your name? | Avani Tanna, Marius Fleischer, Aditya Gulati |
4 | Predicting Earnings Surprises from Conference Call Transcripts with Deep Learning | Ross Koval |
5 | VIP-MM: Viral Protein Mutation Machine | Zhenqiao Song |
6 | Can You Take A Punch?: A Resiliency Survey Against Adversarial Attacks of Deep Models | Aaron Jimenez |
7 | On Neural Ordinary Differential Equations for Natural Language Tasks | Danqing Wang |
8 | Towards Safe and Trustworthy AI Systems with FACT | Alex Mei |
9 | Object Detection using RetinaNet | Brian Lee, Punnal Ismail Khan, Kimia Afshari |
10 | Calculation versus Intuition: Contrasting Decision-Making Processes in Chess AI | Lucas Nguyen |
11 | A Hindsight based Backward & Forward Exploration Strategy for State Space Search Problems. | Vihaan Akshaay Rajendiran |
12 | Machine Predictions on Pretrail Release | Lei Yue |
13 | Dense Passage Retrieval: QANTA BERT | Zihan Ma, Qiming Wu, Shi Bu |
14 | Optimizing Transformer for Low-Resource Yakut-Russian NMT | Kyle Wong |
15 | It is All About Attention | Alfonso Amayuelas Fernandez, Deepak Nathani, Krushna Shah |
16 | Deep Learning based Movie Recommendation System | Pranjali Jain, Vinothini Gunasekaran, Apoorva Jakalannanavar Halappa Manjula |
17 | Visualizing Text with Diffusion Models | Bairu Hou, Yujian Liu, Jiabao Ji |
18 | Deployment efficient reward free RL with linear representation. | Dan Qiao |
19 | Attention for Brain Network Analysis | Christos Zangos |
20 | Agile Compressed Time-Through Back-Propagation Spiking Neural Network for edge intelligence | Boxun Xu |
21 | Neuroformer: A Transformer Framework for Multimodal Neural Data Analysis | Antonis Antoniades |
22 | Improving Diffusion-Language-Model sampling speed with ordinary differential equations | Xianjun Yang |
23 | Differential Privacy in Survival Analysis | Esha Singh |
24 | Hierarchical Representation Learning for Network Data | Danish Ebadulla, Harsha Miryala, Taanya Anand, Satyandra Guthula |
25 | Zero-shot Dense Retrieval with Contrastive Dual Learning | Shanxiu He |
26 | Cross Lingual Transfer Learning for Automatic Speech Recognition | Shinda Huang, Nawel Alioua, Rasta Tadayon Tahmasebi |
27 | Towards Sanskrit-English NMT | Rajan Saini |
28 | Constrained Text Generation from the Pretrained Embedding Space | Wenda Xu |
29 | Parameter- and Memory-Efficient Language Modeling with Ladder-Side Network | Kangyou Yu, Weizhi Wang |
30 | On Recent Progress in Long Sequence Modeling | Siqi Ouyang |
31 | Trajectory Retrieval for Learning-Free Efficient Stable Diffusion | Kexun Zhang |