UNSW Computer Science and Engineering’s Post

From the power of friendship to recovering from disasters, from the secrets of quantum computing to how we’re going to negotiate laws in space, and so much more, join us to hear from UNSW researchers as they present a series of short talks for inquisitive minds on progress; discussing where humanity is headed next. Above & Beyond 🔗 https://lnkd.in/gyAuF6Mj Where Are We Heading? 🔗 https://lnkd.in/gi-yDCVR

I'm currently recruiting at postdoctoral and research programmer (Honours or Masters in Computer Science) levels for a project on automated software verification(model checking), including of smart contracts. Key skills required are functional programming (Haskell), formal methods and/or Solidity. For details, see https://lnkd.in/g_CWA22H https://lnkd.in/gRfr_N-B

Happy to share our AI4Science research on Material Science Knowledge Graph has been accepted for the NeurIPS 2024 Main Track! This work is a joint effort among UNSW SPREE , UNSW Computer Science and Engineering , Microsoft Research and GreenDynamics. Title: Construction and Application of Materials Knowledge Graph in Multidisciplinary Materials Science via Large Language Model ## 🌟 Highlights of our research: - MKG: Now the world's largest and most accurate Knowledge Graph for material science - Innovative pipeline using our 👩🔬Science Large Languge Model (🦘DARWIN https://lnkd.in/gBMassyV) for precise annotation, data extraction, and information traceability - Comprehensive ontology with labels like Name, Formula, and Application, enhancing data usability and integration - Network-based algorithms for efficient link prediction, reducing reliance on traditional experimental methods - Practical applications demonstrated in renewable sector(photovoltaics☀️, battery🔋 and catalyst🧪) material discovery This structured approach not only streamlines materials research but also paves the way for more sophisticated materials knowledge graphs. ## 👥 Meet the Team: Yeyan Peng (Master Student seeking PhD opportunities), Jie Ren, Shaozhou Wang, Yuwei Wan, Imran Razzak, Haofeng Wang, Bram Hoex, Tong Xie, Wenjie Zhang ## 🔗 Learn More: GitHub: https://lnkd.in/g6GEE6wT Preprint: https://lnkd.in/gjWdqTcj Special thanks to National Computational Infrastructure and Pawsey Supercomputing Research Centre for their invaluable HPC support! UNSW AI Institute #AI4Science #MaterialScience #NeurIPS2024 #UNSW #ResearchBreakthrough #largelanguagemodels

UNSW long service awards - Nadine and I both 25 years, and Salil 20 years, 70 years of service at UNSW Computer Science and Engineering. Nadine MarcusSalil Kanhere UNSW Computer Science and Engineering

Rather excitingly, my talk to TEDxUNSW on the meaning of sound is now live. This was a great deal of fun indeed! I am very grateful to all involved - to the organisers for their wonderful invitation, and to the audience for the patience and good humour. UNSW Computer Science and Engineering UNSW https://lnkd.in/gm3tmdCV

Thrilled to share that our paper "CLAP4CLIP: Continual Learning with Probabilistic Finetuning for Vision-Language Models" has been accepted to NeurIPS 2024 ! 🎉 👉 We show that continual fine-tuning of the #CLIP model (based on either prompts or adapters) leads to domain mismatch between the visual and the textual space, deteriorating its cross-modal retrieval performance. We propose variational modelling of the visual-guided text features to resolve this. 👉 To mitigate the forgetting of past task knowledge in continual finetuning of the #CLIP model, we simply exploit the readily available hand-crafted textual prompts for weight initialization and function-space regularization of the downstream model parameterizing the variational posterior of a task. Interestingly, this also helps bridge the #stability gap in continual finetuning! 👉 Checkout our pre-print: https://lnkd.in/gMj-civD 👉 And code: https://lnkd.in/gDr6scFJ A big thanks to my supervisors Lina Yao and Dong Gong for being supportive throughout. Happy to discuss any feedback! #NeurIPS2024