Peng Kang

Experience

• 

  Northwestern University

  Sept 2018 - now

  • CS 449: Deep Learning @ Northwestern University, Spring 2023• CS 349: Machine Learning @ Northwestern University, Fall 2022, Spring 2022, Winter 2022, Winter 2023• CS 348: Intro to Artifical Intelligence @ Northwestern University, Summer 2022, Fall 2021Lead a team of around 6~8 teaching assistants/peer mentors; prepare course materials such as assignments, exams, and handouts; hold regularly scheduled office hours to advise and assist around 150~200 students; answer students’ questions online through Campuswire; initiate, facilitate, and moderate classroom discussions; grade the students’ assignments and evaluate the overall performance. Show less Recommender Systems:• Innovated a hierarchical gating network (HGN), integrated with the Bayesian Personalized Ranking (BPR) to capture both the long-term and short-term user interests.• Created a gated attentive-autoencoder (GATE) model, which is capable of learning fused hidden representations of items’ contents and binary ratings, through a neural gating structure.• Proposed recommendation systems based on variational autoencoders and graph convolutional neural networks.• Demonstrated unmatched performance on real-world datasets, setting the benchmark for state-of-the-art recommendation systems. Show less Imaging Processing and Deep LearningAttentional Text Matting:• Pioneered the development of a novel text matting pipeline, introducing the innovative capability to extract WordArts seamlessly from image backgrounds.• Spearheaded the creation of an image synthesis engine and curated a comprehensive dataset featuring 46,289 images, annotated to empower seamless learning and evaluation on text matting.• Orchestrated an exhaustive array of quantitative and qualitative experiments, spanning both synthetic datasets and real-world images, achieving a state-of-the-art performance in the domain of text matting.Bio-inspired Robust Object Recognition:• Leveraged recent breakthroughs in Artificial Neural Networks (ANNs) and Spiking Neural Networks (SNNs) to pioneer the development of bio-inspired hierarchical ANN-SNN models, revolutionizing the landscape of robust object recognition.• Innovated by drawing from concepts such as denoising autoencoders, residual skip connections, and gating mechanisms, unveiled three cutting-edge ANN-SNN variant models that enhance object recognition robustness.• Evidenced the exceptional performance of the ANN-SNN models by surpassing previous SNN benchmarks, establishing a new state-of-the-art standard in the realm of robust object recognition.Energy-efficient Object Recognition:• Brought spiking neurons into Geoffrey Hinton's GLOM model, resulting in the groundbreaking Spiking-GLOMs for object recognition.• Obtained comparable object recognition accuracy and interpretability to the GLOM model while reducing energy consumption, with Spiking-GLOMs proving up to 40x more energy-efficient. Show less

  • Teaching Assistant

    Sept 2021 - now

  • Deep Learning Researcher

    Sept 2018 - now

  • Deep Learning Researcher

    Sept 2018 - now

  • Deep Learning Researcher

    Sept 2018 - now
• 

  Dolby Laboratories

  Jun 2023 - Sept 2023

  Machine Learning and Imaging Research Intern

  • Developed the industry’s first pipeline to generate haptic stiffness senses of objects in the images.• Employed the material segmentation, object segmentation, vision-language model CLIP, and large language model -- GPT-3.5 to construct the pipeline. • Showcased precise stiffness recognition of objects within images through the extensive prompt engineering, elevating the system to a human-level performance standard.