Blog series / learning path
Deep Learning from First Principles
A paced route through the parts of deep learning that keep reappearing in real systems.
01 / 01 CHAPTERSIN PROGRESS
Prerequisites / related Atlas
ATLAS
Mathematics & StatisticsThe quantitative language for learning, estimation, geometry, and uncertainty.FoundationsProgramming & SoftwareThe software abstractions that make experiments reproducible and systems maintainable.FoundationsComputer ScienceAlgorithms, systems, networking, and concurrency beneath intelligent machines.FoundationsElectronics & Embedded SystemsReal-time hardware, sensors, and actuators where computation meets the physical world.FoundationsMachine LearningLearning useful structure from data, objectives, and feedback.Artificial IntelligenceDeep LearningDifferentiable representations optimized from data and objectives.Artificial IntelligenceComputer VisionTurning images and video into structured state, identity, and affordances.Artificial IntelligenceLanguage Models & LLMsLanguage models as predictors, representations, and interfaces for tools.Artificial IntelligenceMultimodal Learning & VLMsAligning visual and language representations so models can ground symbols in scenes.Artificial IntelligenceReinforcement LearningPolicies learned through rewards, interaction, and long-horizon consequences.Artificial IntelligenceAI Agents & Agentic SystemsModels, tools, memory, and feedback loops organized around goals.Artificial IntelligenceRobot Mechanics, Kinematics & ControlThe physical models and feedback laws that make motion predictable and controllable.RoboticsSensing & Robot PerceptionSensors and estimators that turn imperfect measurements into usable state.RoboticsSLAM & NavigationLocalization, mapping, planning, and control coordinated under motion constraints.RoboticsRobot Software Systems & CommunicationMiddleware, communication, and runtime structure for composing robot systems.RoboticsManipulation & GraspingContact, grasp, and motion planning for robots that handle the physical world.RoboticsLegged & Humanoid RoboticsBalance, locomotion, and whole-body motion under contact and embodiment.RoboticsRobotics Simulation & ToolchainsSimulation environments and digital twins for testing learning and control safely.RoboticsEmbodied Perception & Spatial IntelligencePerception that is situated, active, and useful for choosing what to do next.Embodied IntelligenceEmbodied NavigationWayfinding and exploration grounded in language, geometry, and action.Embodied IntelligenceEmbodied Manipulation & DexterityLong-horizon, contact-rich manipulation organized around affordances and feedback.Embodied IntelligenceHumanoid & Whole-Body IntelligenceWhole-body policies and contact-aware skills for humanoid agents.Embodied IntelligenceRobot Foundation Models & VLAVision-language-action models that connect perception, language, and robot policy.Embodied IntelligenceWorld Models & Embodied MemoryPredictive state, dynamics, and memory for imagining what an action may cause.Embodied IntelligenceEmbodied Agents & Long-Horizon AutonomyAgents that plan, remember, recover, and act over extended physical tasks.Embodied IntelligenceRobot Learning & Policy LearningPolicies learned from demonstrations, rewards, and repeated interaction with a robot.Embodied IntelligenceEmbodied Data, Simulation & Sim-to-RealDemonstrations, synthetic worlds, and transfer procedures that make robot learning scalable.Embodied IntelligenceBLOG
从反向传播到一次可复现的梯度实验用一个最小的计算图，把链式法则、自动微分和实验记录放在同一张工作台上。deep-learning · Deep Learning from First Principles · Deep Learning · 06 minPAPERS
Deep SORTReading note in progress.ICIP · 2017 · multi-object-trackingMeMOTRReading note in progress.ICCV · 2023 · multi-object-trackingMOTRReading note in progress.ECCV · 2022 · multi-object-trackingMOTRv2Reading note in progress.CVPR · 2023 · multi-object-trackingSORTReading note in progress.ICIP · 2016 · multi-object-trackingTrackFormerReading note in progress.CVPR · 2021 · multi-object-trackingPROJECTS
LoneRoverA complete ROS 2 navigation stack for an Ackermann vehicle, connecting state estimation, planning, control, and safety on a real competition platform.Real-world robotics · ROS 2 · EKF · Nav2 · MPPI · RPPcorrMOTRA MOTR-oriented research system that separates geometry/motion, appearance identity, and existence/lifecycle into explicit state domains.Research prototype · MOTR · DETR · TrackPool · StateTSMA Temporal Shift Module pipeline for real-time gesture recognition, with data preparation, training configuration, checkpoint loading, and webcam inference.Temporal vision / HRI · TSM · Video · HRI · PyTorchNo matching records.
Blog series / learning path
A paced route through the parts of deep learning that keep reappearing in real systems.
Prerequisites / related Atlas