Aerospike’s Post

If there's one thing Takahiro Nakadai has that is unmatched, it is his passion for Aerospike! Here's Tak presenting at the satellite stage of db tech showcase, on the origins of Aerospike and our use cases. #dbts Aveekshith Bushan Tomo Kikuchi Honon Matsuo Kristle Chia

Learn the workflow from the pros!

Check out the latest episode of the Stantec.io podcast where we discuss advanced technology, drones, and reality mesh!

Check out the latest episode of the Stantec.io podcast where we discuss how advanced technology, drones, and reality mesh!

Pay attention to the detail ;-) . It is not about the best possible image, it is about reliable detection. Direct distance in addition to standard IR image makes the system more reliable.

Check out the latest episode of the Stantec.io podcast where we discuss advanced technology, drones, and reality mesh!

NEW AllThingsXR episode is up! Dive into the world of #XR and #Robotics with Niko, CEO of Rerun. With years of expertise in computer vision and machine learning, Nikolaus West is reshaping data visualization for multimodal data via Rerun. https://lnkd.in/em5zZ3ac

Introducing AROUND, Techtree Innovation's spatial digital twin platform that implements high-resolution, high-precision, 3D spatial information construction and visualization technology. Various physical laws and effects such as collision, reflection, gravity, and light are also implemented, and the accuracy and resolution are within 5cm by unifying high-altitude, low-altitude, and ground resolution, and are customized and supplied in four solutions: AROUND.city, AROUND. real, AROUND. sim, and AROUND. map #3d #spatial #geospatial #digitaltwin #drone #photogrammetry #unrealengine #unreal #metaverse https://lnkd.in/gUyGtxYz

I'll be attending ICML in Vienna next week. Our team has a paper presented at the conference, and I'm excited to reconnect with friends and meet new colleagues. Feel free to reach out if you'd like to chat. Below are some of my recent projects, and you can find more details at https://lnkd.in/gpj5ChfT. At ICML, we are presenting Is "flatness" really what drives generalization in Sharpness-Aware Minimization (SAM)? Forget sharpness; we think the training dynamics are key. 📄 Read the paper here: https://lnkd.in/gFvnGWNH. Joint work with Ankit Vani, Frederick Tung, Gabriel Oliveira and Hossein Sharifi-Noghabi SAM is traditionally seen as a method to minimize sharpness to improve generalization, but a causal link between flatness and generalization is unclear. Prior work shows they can be uncorrelated, and practical tricks like m-sharpness deviate from SAM's original theory. We propose a new perspective on SAM's training dynamics, viewing it as a variant of the iterative process of forgetting and relearning. In this view, perturbations in SAM act as a mechanism for forgetting undesirable model biases. Relearning can help humans integrate concepts more effectively with their existing knowledge. Similarly, perturbed forgetting in SAM estimates update directions by leveraging global structures after discarding shortcuts the model may adopt for individual samples. Standard SAM perturbations can sometimes inhibit beneficial forgetting by discarding some while exacerbating other biases. We introduce the output bias forgetting (OBF) perturbation, which mitigates this. OBF outperforms standard perturbations in SAM, GSAM, and ASAM. Finally, we show that an information-theoretic measure of perturbed forgetting can correlate more strongly with generalization than flatness! Check out the paper for more results and discussion. Our code is available at https://lnkd.in/gt9r8aCj.

🤖 Watch our CTO’s talk on RAI, our agent-based AI Framework for ROS 2! ROSCon is over, but the great news is that the insights shared during the conference are available online. Adam Dabrowski gave a short and informative introduction to RAI so be sure to check it out here: https://lnkd.in/dKiDJqH6. ᯓ★ To start your first RAI project, head to GitHub: https://lnkd.in/dhvy9jxe Open Robotics #ROSCon #ROS #simulation #robotics