Abstract: Our aim is to learn to solve long-horizon decision-making problems in complex robotics domains given low-level skills and a handful of demonstrations containing sequences of images. To this ...

Abstract: The performance of current Scene Graph Generation models is severely hampered by some hard-to-distinguish predicates, e.g., “woman-on/standing on/walking on-beach” or “woman-near/looking ...

The performance of current Scene Graph Generation (SGG) models is severely hampered by hard-to-distinguish predicates, e.g., "woman-on/standing on/walking on-beach". As general SGG models tend to ...

Classic query optimization techniques, including predicate pushdown, are of limited use for machine learning inference queries, because the user-defined functions (UDFs) which extract relational ...

Imagine having to narrow down a video of passing vehicles captured on a traffic camera to red SUVs that are exceeding the speed limit. Such queries over images, videos and text are an increasingly ...

This feature enables the use of the flag "learnMode" in the configuration to be used to learn predicates in the template that best characterize the traces. The learnMode's are as follows: learnMode=0: ...

I want to confirm one thing about backpropagation adjustment. In the RC layer of GPT architecture, the responsibility error at each predicate is distributed/allocated ...

I want to confirm one thing about backpropagation adjustment. In the RC layer of GPT architecture, the responsibility error at each predicate is distributed/allocated ...

Agent can be warm-started with models by pointing to files using args. E.g., to load interaction rules stored in model_demo_level2.py, operators in operators.py, predicates in predicates.py, and to ...