🚀 Welcome to the latest Marvik Digest 🚀
Last month we covered some interesting stories involving new advances in Conversational AI, NVIDIA’s latest achievement on neural graphics primitives, the challenges of Real-Time Bidding in digital advertising, and more.
➡️ Want us to cover a specific topic? Reach out to [email protected] to send us your suggestions.
Advances in Conversational AI with BlenderBot 2.0
🚀 New advances in #ConversationalAI 🚀
Meta AI Research has built and open-sourced #BlenderBot 2.0, the first chatbot that can at the same time build long-term memory it can continually access, search the internet for timely information, and have high-level conversations on almost any topic.
🟢 Main highlights:
📌 Significant update to the original BlenderBot – it’s better at conducting longer, more knowledgeable, and factually consistent conversations over multiple sessions.
📌 Uses a model based on Facebook’s #RetrievalAugmentedGeneration — it allows to generate dialogue responses that include knowledge beyond that contained in the conversation itself.
📌 The model takes pertinent information extracted during conversation and stores it in a long-term memory so it can then leverage this knowledge in ongoing conversations that may continue for days, weeks, or even months.
📌 During conversation, the model can generate contextual internet search queries, read the results, and incorporate that information when responding to people’s questions and comments.
📌 The complete model, code, evaluation setup, and two new conversational data sets used to train the model were released → incentive to reproduce and advance on research.
➡️ More on this new model here: https://bit.ly/3bIO98y
NVIDIA’s Instant NeRF gets awarded
“Instant Neural Graphics Primitives with a Multiresolution Hash Encoding” demonstrates near-instant training of neural graphics primitives on a single #GPU for multiple tasks.
🟢 Main takeaways:
📌 Reduces costs of training neural graphics primitives, with a new input encoding that allows the use of a smaller network without compromising on quality
📌 Achieves combined speedup of several orders of magnitude
📌 Training of high-quality neural graphics primitives in seconds
📌 Rendering in tens of milliseconds at 1920×1080
📌 Source code and other resources available
SimpleChains.jl for small neural networks
When using #ML general frameworks, you usually have to sacrifice performance for specific tasks.
📌 Started as a solution for scientific machine learning (#SciML) in healthcare data analytics
📌 Not universally applicable, but useful for specific cases
➡️ Click here to learn more about SimpleChains.jl: https://bit.ly/3psusos
Real-Time Bidding in digital advertising
In our latest blog post, our #datascientist María Sofía Pérez Casulo gives a walkthrough of the Real-Time Bidding (#RTB) process in digital advertising and its challenges. She discusses the main challenges of RTB and presents alternative strategies to optimize its bidding price, a key aspect to improve campaign performances and increase profits.
“Real-time bidding (RTB) is a challenging matter, and definitely an excellent opportunity to apply deep-learning techniques. As the industry becomes more mature, more challenges and fruitful research opportunities arise”.
👉 Visit our blog for the full story https://bit.ly/3Ttra23