Explore

Reasoning/Symbolic: A Survey

Table

Table

Name

Journals/Conferences

Year

Categories

Issues

Motivations

Contributions

Potential ideas

URL

⁠

Solving olympiad geometry without human demonstrations | Nature

⁠

A simple neural network module for relational reasoning

- Symbol grounding problem. - Not robust to small task. - Input variations. - data-poor problems → sparse but complex relations.

- Relation networks (RNs): single function to compute/score all relations. - RN-augmented architecture.

Deep Logic: Joint Learning of Neural Perception and Logical Reasoning

- Hard to reach global optima. - Not designed to deal with semantic data (images & text).

- Joint learning framework.

- Build Logical Network consists of: + Multi-nomial(x ; h_k). + h_k: output of BiLSTM (model dependencies between different layers). + The representations operates the finite recursion to feed through the logical gate multiple times. - Training (Optimization Stage): + Optimize encoder, fix logical network: + Optimize logical network, fix encoder:

- Disentangle to reasoning. - Disentangled representations as symbolics. - The network is not adaptive, the logical nodes are fixed. There should be an approach to reuse the logical nodes (via LSTM/RNN) or Regularization like Pruning/Quantization etc.

Bayesian Inverse Contextual Reasoning for Heterogeneous Semantics-Native Communication

Trading off Utility, Informativeness, and Complexity in Emergent Communication.

Human languages are shaped by task-general communicative constraints.

Leverages generative adversarial network.

Optimize a tradeoff between maximizing utility, informativeness vs. minimizing complexity.

Minimize the complexity of signal → explainable.

Disentanglement of c into invariant and non-invariant signal.

Reasoning from the invariant signal.

Alexa Arena: A User-Centric Interactive Plaform for Embodied AI

Human-like systematic generalization through a meta-learning neural network

SwiftSage: A Generative Agent with Fast and Slow Thinking for Complex Interactive Tasks

Iterated Learning Improves Compositionality in Large Vision-Language Models

Nethack is Hard to Hack

Grounding Neural Inference with Satisfiability Modulo Theories

Active Reasoning in an Open-World Environment

Egocentric Planning for Scalable Embodied Task Achievement

Efficient Symbolic Policy Learning with Differentiable Symbolic Expression

What’s Left? Concept Grounding with Logic-Enhanced Foundation Models

A social path to human-like artificial intelligence

When the world is large enough that the initial behavior distribution covers only a relatively tiny subspace.

Natural Intelligence emerges at multiple scales in networks of interacting agents via collective living, social relationships and major evolutionary transitions.

Learning quality depends on the richness and size of the dataset.

Bottleneck in AI shifting from data assimilation to novel data generation.

Exploration and Exploitation can be synergistic instead of antagonistic.

Compounding Innovation: Exploitation drives exploration → appropriate direction.

A Dual Representation Framework for Robot Learning with Human Guidance

Neurosymbolic AI for Reasoning Over Knowledge Graphs: A Survey

Scene-Driven Multi-modal Knowledge Graph Construction for Embodied AI

Semantic HELM: A Human-Readable Memory for Reinforcement Learning

Describe, Explain, Plan and Select: Interactive Planning with Large Language Models Enables Open-World Multi-Task Agents

Leveraging Symbolic Knowledge Bases for Commonsense Natural Language Inference Using Pattern Theory

State2Explanation: Concept-Based Explanations to Benefit Agent Learning and User Understanding

Read and Reap the Rewards: Learning to Play Atari with the Help of Instruction Manuals

3D Concept Learning and Reasoning from Multi-View Images

Large Language Models Are Neurosymbolic Reasoners

Interactive Visual Reasoning under Uncertainty

EgoTV : Egocentric Task Verification from Natural Language Task Descriptions

Interpretable Imitation Learning with Symbolic Rewards

EqMotion: Equivariant Multi-agent Motion Prediction with Invariant Interaction Reasoning

Interpretable and Explainable Logical Policies via Neurally Guided Symbolic Abstraction

There are no rows in this table

⁠

Want to print your doc?
This is not the way.

Try clicking the ··· in the right corner or using a keyboard shortcut (

CtrlP

) instead.