Compared with static knowledge graphs, temporal graphs (tKG), which can capture the evolution and change of information over time, are more realistic general. However, due to complexity that notion time introduces learning rules, an accurate graph reasoning, e.g., predicting new links between entities, is still a difficult problem. In this paper, we propose TILP, differentiable framework for logical rules learning. By designing constrained random walk mechanism introduction operators, ensure...

Conventional embedding-based models approach event time prediction in temporal knowledge graphs (TKGs) as a ranking problem. However, they often fall short capturing essential relationships such order and distance. In this paper, we propose TEILP, logical reasoning framework that naturaly integrates elements into graph predictions. We first convert TKGs (TEKG) which has more explicit representation of term nodes the graph. The TEKG equips us to develop differentiable random walk prediction....

Incipient faults in power distribution systems potentially lead to catastrophic failures. Detection of incipient contributes proactive fault management and predictive maintenance, which effectively improves supply reliability. Since the are infrequent transient, few samples can be procured real applications. In this paper, a detection method based on human-level concept learning (HLCL) is proposed address problem. The contains two steps: waveform decomposition (HLWD) hierarchical...

In this paper, we address hypothesis testing in a distributed network of nodes, where each node has only partial information about the State World (SotW) and is tasked with determining which hypothesis, among given set, most supported by data available within node. However, due to node's limited perspective SotW, individual nodes cannot reliably determine independently. To overcome limitation, must exchange via an intermediate server. Our objective introduce novel lossy semantic...

In this paper, we propose an advancement to Tarskian model-theoretic semantics, leading a unified quantitative theory of semantic information and communication. We start with description inductive logic probabilities, which serve as notable tools in development the proposed theory. Then, identify two disparate kinds uncertainty communication, that physical content, present refined interpretations measures, conclude proposing new measure for content-information entropy. Our proposition...

Inspection robots are popularized in substations due to the lack of personnel for operation and maintenance. However, these inspection remain at level perceptual intelligence, rather than cognition intelligence. To enable a robot automatically detect defects power equipment, object recognition is critical step because criteria infrared diagnosis vary with types equipment. Since this task not big-sample learning problem, prior knowledge needs be added improve existing methods. Here, an model...

The power distribution system's fault root cause classification is an important but challenging problem. Traditional classifiers fail to achieve high accuracy and good generalization performance due data insufficiency. A large volume of unlabeled available, which can be utilized improve performance. This paper proposes a novel classifier called Robust Semi-Supervised Prototypical Network (RSSPN) based on architecture semi-supervised learning address this issue. proposed method mine...

In order to improve the accuracy of power equipment recognition, an image recognition method based on Mask RCNN and Bayesian Context Network is proposed. The two-layer network contains R-CNN as first layer, which gives preliminary results, second utilizes context information correct results. designed take relationship type, size, spatial location between objects into account. Experiments images in substation show that proposed outperforms R-CNN, Multi-task Cascades Fully Convolutional...

Inductive logic reasoning is a fundamental task in graph analysis, which aims to generalize patterns from data. This has been extensively studied for traditional representations, such as knowledge graphs (KGs), using techniques like inductive programming (ILP). Existing ILP methods assume learning KGs with static facts and binary relations. Beyond KGs, structures are widely present other applications procedural instructions, scene graphs, program executions. While beneficial these...

Recent advancements in large language models (LLMs) have significantly enhanced their capacity to aggregate and process information across multiple modalities, enabling them perform a wide range of tasks such as multimodal data querying, tool usage, web interactions, handling long documents. These capabilities pave the way for transforming LLMs from mere chatbots into general-purpose agents capable interacting with real world. This paper explores concept using model core component an...

In this paper, we address the problem of lossy semantic communication to reduce uncertainty about State World (SotW) for deductive tasks in point communication. A key challenge is transmitting maximum information with minimal overhead suitable downstream applications. Our solution involves maximizing content within a constrained bit budget, where SotW described using First-Order Logic, and informativeness measured by usefulness transmitted reducing perceived receiver. Calculating requires...

Enhancing the reasoning capabilities of large language models (LLMs) remains a key challenge, especially for tasks that require complex, multi-step decision-making. Humans excel at these by leveraging deliberate planning with an internal world model to simulate potential outcomes various actions. Inspired this, we propose novel framework LLMs, referred as Structure-aware Planning Accurate World Model (SWAP). Unlike previous approaches rely solely on Chain-of-Thought (CoT) in natural...

While large language models (LLMs) have demonstrated remarkable reasoning capabilities, they are not without their flaws and inaccuracies. Recent studies introduced various methods to mitigate these limitations. Temporal (TR), in particular, presents a significant challenge for LLMs due its reliance on diverse temporal expressions intricate contextual details. In this paper, we propose TG-LLM, new framework towards language-based TR. To be specific, first teach LLM translate the context into...

Large Language Models (LLMs) have shown superior capability to solve reasoning problems with programs. While being a promising direction, most of such frameworks are trained and evaluated in settings prior knowledge task requirements. However, as LLMs become more capable, it is necessary assess their abilities realistic scenarios where many real-world open-ended ambiguous scope, often require multiple formalisms solve. To investigate this, we introduce the wild, an LLM tasked problem unknown...

Causal reasoning (CR) is a crucial aspect of intelligence, essential for problem-solving, decision-making, and understanding the world. While large language models (LLMs) can generate rationales their outputs, ability to reliably perform causal remains uncertain, often falling short in tasks requiring deep causality. In this survey, we provide comprehensive review research aimed at enhancing LLMs reasoning. We categorize existing methods based on role LLMs: either as engines or helpers...

Translating natural language sentences to first-order logic (NL-FOL translation) is a longstanding challenge in the NLP and formal literature. This paper introduces LogicLLaMA, LLaMA-7B model fine-tuned for NL-FOL translation using LoRA on single GPU. LogicLLaMA capable of directly translating into FOL rules, which outperforms GPT-3.5. also equipped correct rules predicted by GPT-3.5, can achieve similar performance as GPT-4 with fraction cost. correction ability was achieved novel...

Conventional embedding-based models approach event time prediction in temporal knowledge graphs (TKGs) as a ranking problem. However, they often fall short capturing essential relationships such order and distance. In this paper, we propose TEILP, logical reasoning framework that naturally integrates elements into graph predictions. We first convert TKGs (TEKG) which has more explicit representation of term nodes the graph. The TEKG equips us to develop differentiable random walk prediction....

Inductive logic reasoning is one of the fundamental tasks on graphs, which seeks to generalize patterns from data. This task has been studied extensively for traditional graph datasets such as knowledge graphs (KGs), with representative techniques inductive programming (ILP). Existing ILP methods typically assume learning KGs static facts and binary relations. Beyond KGs, structures are widely present in other applications video instructions, scene program executions. While also beneficial...