University of British Columbia
AI Insights
- Reranking: A selection strategy that uses retrieved options to select the best candidate for code generation. (ML: 0.97)
- Closed-source models may not benefit as much from PKG-based retrieval due to their high no-retrieval baselines and potential risks associated with irrelevant or contradictory context. (ML: 0.96)
- Re-ranking can further improve performance by selecting the best candidates among retrieved options. (ML: 0.95)
- Programming Knowledge Graph (PKG): A knowledge graph specifically designed for programming tasks, which captures relationships between concepts and entities in a program. (ML: 0.95)
- Topic-level analysis reveals that PKG outperforms BM25 across all topics, especially in those where lexical matching is less reliable. (ML: 0.94)
- The effectiveness of PKG-based retrieval depends on various factors, including the representation method used, the model's adaptability to different problem topics, and the selection strategy employed. (ML: 0.93)
- The effectiveness of PKG-based retrieval depends on various factors, including representation method, model adaptability, and selection strategy. (ML: 0.93)
- Retrieval-augmented code generation systems can benefit from the use of Programming Knowledge Graphs (PKG) to improve performance and robustness. (ML: 0.92)
- Representation method: The way in which the PKG is represented, such as using row/Q&A representations, function-level representations, or block-level representations. (ML: 0.91)
- Retrieval-augmented code generation systems: Systems that use retrieval to augment code generation capabilities. (ML: 0.88)
- The use of PKG-based retrieval can improve performance and robustness in code generation systems. (ML: 0.85)
Abstract
Large Language Models (LLMs) excel at code generation but struggle with complex problems. Retrieval-Augmented Generation (RAG) mitigates this issue by integrating external knowledge, yet retrieval models often miss relevant context, and generation models hallucinate with irrelevant data. We propose Programming Knowledge Graph (PKG) for semantic representation and fine-grained retrieval of code and text. Our approach enhances retrieval precision through tree pruning and mitigates hallucinations via a re-ranking mechanism that integrates non-RAG solutions. Structuring external data into finer-grained nodes improves retrieval granularity. Evaluations on HumanEval and MBPP show up to 20% pass@1 accuracy gains and a 34% improvement over baselines on MBPP. Our findings demonstrate that our proposed PKG approach along with re-ranker effectively address complex problems while maintaining minimal negative impact on solutions that are already correct without RAG. The replication package is published at https://github.com/iamshahd/ProgrammingKnowledgeGraph
Large Language Models (LLMs) excel at code generation but struggle with complex problems. Retrieval-Augmented Generation (RAG) mitigates this issue by integrating external knowledge, yet retrieval models often miss relevant context, and generation models hallucinate with irrelevant data. We propose Programming Knowledge Graph (PKG) for semantic representation and fine-grained retrieval of code and text. Our approach enhances retrieval precision through tree pruning and mitigates hallucinations via a re-ranking mechanism that integrates non-RAG solutions. Structuring external data into finer-grained nodes improves retrieval granularity. Evaluations on HumanEval and MBPP show up to 20% pass@1 accuracy gains and a 34% improvement over baselines on MBPP. Our findings demonstrate that our proposed PKG approach along with re-ranker effectively address complex problems while maintaining minimal negative impact on solutions that are already correct without RAG. The replication package is published at https://github.com/iamshahd/ProgrammingKnowledgeGraph
Why we are recommending this paper?
Due to your Interest in Object Oriented Programming
This paper explores techniques for improving code generation, a key area of interest given your focus on programming paradigms and language design. The use of knowledge graphs to augment LLMs directly addresses the challenges of complex code generation problems.
Due to your Interest in Object Oriented Programming
This paper explores techniques for improving code generation, a key area of interest given your focus on programming paradigms and language design. The use of knowledge graphs to augment LLMs directly addresses the challenges of complex code generation problems.
University of Padua
AI Insights
- Relative completeness relies on the rules (basic), (seq), (choice), (iter), (cons), and (join). (ML: 0.94)
- Relative completeness: The ability to derive a valid Hoare triple from its semantic properties. (ML: 0.93)
- Simple monoid interpretation: An interpretation where intervals are combined using the union operation. (ML: 0.89)
- Hoare logic: A formal system for specifying and reasoning about programs with interval-based semantics. (ML: 0.88)
- The rule (iter) is essential for relative completeness, as demonstrated by Example 3.5. (ML: 0.86)
- Hoare logic can be formulated in two ways using either the irreducible or simple monoid interpretation. (ML: 0.84)
- Irreducible monoid interpretation: An interpretation where intervals are not combined using the union operation. (ML: 0.84)
- APPL: A proof system for programs with interval-based semantics. (ML: 0.84)
- The APPL proof system for programs with interval-based semantics is relatively complete when using the irreducible monoid interpretation. (ML: 0.81)
- The (join) rule is not essential for the relative completeness of standard Hoare logic. (ML: 0.78)
Abstract
We introduce APPL (Abstract Program Property Logic), a unifying Hoare-style logic that subsumes standard Hoare logic, incorrectness logic, and several variants of Hyper Hoare logic. APPL provides a principled foundation for abstract program logics parameterised by an abstract domain, encompassing both existing and novel abstractions of properties and hyperproperties. The logic is grounded in a semantic framework where the meaning of commands is first defined on a lattice basis and then extended to the full lattice via additivity. Crucially, nondeterministic choice is interpreted by a monoidal operator that need not be idempotent nor coincide with the lattice join. This flexibility allows the framework to capture collecting semantics, various classes of abstract semantics, and hypersemantics. The APPL proof system is sound, and it is relatively complete whenever the property language is sufficiently expressive. When the property language is restricted to an abstract domain, the result is a sound abstract deduction system based on best correct approximations. Relative completeness with respect to a corresponding abstract semantics is recovered provided the abstract domain is complete, in the sense of abstract interpretation, for the monoidal operator.
We introduce APPL (Abstract Program Property Logic), a unifying Hoare-style logic that subsumes standard Hoare logic, incorrectness logic, and several variants of Hyper Hoare logic. APPL provides a principled foundation for abstract program logics parameterised by an abstract domain, encompassing both existing and novel abstractions of properties and hyperproperties. The logic is grounded in a semantic framework where the meaning of commands is first defined on a lattice basis and then extended to the full lattice via additivity. Crucially, nondeterministic choice is interpreted by a monoidal operator that need not be idempotent nor coincide with the lattice join. This flexibility allows the framework to capture collecting semantics, various classes of abstract semantics, and hypersemantics. The APPL proof system is sound, and it is relatively complete whenever the property language is sufficiently expressive. When the property language is restricted to an abstract domain, the result is a sound abstract deduction system based on best correct approximations. Relative completeness with respect to a corresponding abstract semantics is recovered provided the abstract domain is complete, in the sense of abstract interpretation, for the monoidal operator.
Why we are recommending this paper?
Due to your Interest in Object Oriented Programming
This work presents a novel logic for program analysis, aligning with your interest in programming paradigms and design patterns. The focus on abstract properties offers a foundational approach to understanding and reasoning about programs.
Due to your Interest in Object Oriented Programming
This work presents a novel logic for program analysis, aligning with your interest in programming paradigms and design patterns. The focus on abstract properties offers a foundational approach to understanding and reasoning about programs.
University College Cork
AI Insights
- Value misalignment: LLMs may not align with human values, leading to unintended consequences. (ML: 0.99)
- Researchers are working on developing methods to address these issues, including value alignment, model observability, and uncertainty-aware language agents. (ML: 0.98)
- The results of these studies highlight the potential benefits and challenges associated with the use of LLMs. (ML: 0.98)
- Lack of transparency: LLMs are often opaque, making it difficult to understand their decision-making processes. (ML: 0.97)
- Several studies have investigated the use of LLMs in various applications, including natural language processing, text generation, and decision-making. (ML: 0.97)
- Researchers have proposed several methods to address the concerns related to LLMs, including value alignment, model observability, and uncertainty-aware language agents. (ML: 0.97)
- LLM: A type of artificial intelligence model that is trained on large amounts of data to generate human-like responses or perform tasks such as translation, summarization, and question-answering. (ML: 0.96)
- Imagine you're using a super smart computer program that can understand and respond to human language in a way that's almost like talking to another person. (ML: 0.95)
- This is basically what Large Language Models (LLMs) are. (ML: 0.95)
- These methods aim to improve the transparency, safety, and security of LLMs. (ML: 0.94)
- Large Language Models (LLMs) have the potential to revolutionize various industries, but their use also raises several concerns regarding their safety, security, and interpretability. (ML: 0.94)
- Large Language Models (LLMs) are being increasingly used in various applications, including natural language processing, text generation, and decision-making. (ML: 0.94)
- The use of LLMs in various applications has raised several concerns regarding their safety, security, and interpretability. (ML: 0.94)
- Adversarial attacks: LLMs can be vulnerable to adversarial attacks, which can compromise their safety and security. (ML: 0.92)
- They have the potential to revolutionize various industries, but their use also raises several concerns regarding their safety, security, and interpretability. (ML: 0.81)
Abstract
With the development of foundation model (FM), agentic AI systems are getting more attention, yet their inherent issues like hallucination and poor reasoning, coupled with the frequent ad-hoc nature of system design, lead to unreliable and brittle applications. Existing efforts to characterise agentic design patterns often lack a rigorous systems-theoretic foundation, resulting in high-level or convenience-based taxonomies that are difficult to implement. This paper addresses this gap by introducing a principled methodology for engineering robust AI agents. We propose two primary contributions: first, a novel system-theoretic framework that deconstructs an agentic AI system into five core, interacting functional subsystems: Reasoning & World Model, Perception & Grounding, Action Execution, Learning & Adaptation, and Inter-Agent Communication. Second, derived from this architecture and directly mapped to a comprehensive taxonomy of agentic challenges, we present a collection of 12 agentic design patterns. These patterns - categorised as Foundational, Cognitive & Decisional, Execution & Interaction, and Adaptive & Learning - offer reusable, structural solutions to recurring problems in agent design. The utility of the framework is demonstrated by a case study on the ReAct framework, showing how the proposed patterns can rectify systemic architectural deficiencies. This work provides a foundational language and a structured methodology to standardise agentic design communication among researchers and engineers, leading to more modular, understandable, and reliable autonomous systems.
With the development of foundation model (FM), agentic AI systems are getting more attention, yet their inherent issues like hallucination and poor reasoning, coupled with the frequent ad-hoc nature of system design, lead to unreliable and brittle applications. Existing efforts to characterise agentic design patterns often lack a rigorous systems-theoretic foundation, resulting in high-level or convenience-based taxonomies that are difficult to implement. This paper addresses this gap by introducing a principled methodology for engineering robust AI agents. We propose two primary contributions: first, a novel system-theoretic framework that deconstructs an agentic AI system into five core, interacting functional subsystems: Reasoning & World Model, Perception & Grounding, Action Execution, Learning & Adaptation, and Inter-Agent Communication. Second, derived from this architecture and directly mapped to a comprehensive taxonomy of agentic challenges, we present a collection of 12 agentic design patterns. These patterns - categorised as Foundational, Cognitive & Decisional, Execution & Interaction, and Adaptive & Learning - offer reusable, structural solutions to recurring problems in agent design. The utility of the framework is demonstrated by a case study on the ReAct framework, showing how the proposed patterns can rectify systemic architectural deficiencies. This work provides a foundational language and a structured methodology to standardise agentic design communication among researchers and engineers, leading to more modular, understandable, and reliable autonomous systems.
Why we are recommending this paper?
Due to your Interest in Design Patterns
This paper investigates agentic AI systems, a growing area of interest given your interest in programming paradigms and design patterns. The system-theoretic framework provides a structured approach to designing reliable AI systems.
Due to your Interest in Design Patterns
This paper investigates agentic AI systems, a growing area of interest given your interest in programming paradigms and design patterns. The system-theoretic framework provides a structured approach to designing reliable AI systems.
University of Texas at Dallas
AI Insights
- Imagine you're trying to match people with jobs in a company. (ML: 0.96)
- By analyzing these paths or cycles, we can figure out which method is more efficient and fair. (ML: 0.95)
- You have two different ways of doing it, and you want to know which one is better. (ML: 0.95)
- The authors also assume that every disagreement edge is accounted for exactly once in the decomposition, which may not be feasible due to supplier capacities. (ML: 0.92)
- These paths or cycles are like routes that show how each person or job is matched in both methods. (ML: 0.91)
- The main idea is to represent the disagreement set as an auxiliary directed graph and reduce the decomposition problem to two classical graph-theoretic tasks: finding augmenting paths and performing Euler-tour decompositions. (ML: 0.91)
- The paper assumes that the alternating path/cycle decomposition of the disagreement set is unique, which may not be true in many-to-one matching. (ML: 0.90)
- However, in many-to-one matching, the decomposition may admit multiple distinct alternating decompositions, and not all decompositions are feasible due to supplier capacities. (ML: 0.89)
- Definition 1: Alternating Path/Cycle Decomposition of Disagreement Set Definition 2: Feasible Matching in Many-to-One Matching with Capacity Constraint Definition 3: Auxiliary Unbalanced Directed Graph The AP design and the associated inference results from the one-to-one setting continue to apply without modification if a collection of alternating paths and cycles can be identified whose randomized realizations form a feasible matching in M. (ML: 0.88)
- The paper introduces a way to break down the differences between these two methods into smaller pieces called 'alternating paths' or 'cycles'. (ML: 0.83)
- The alternating path/cycle decomposition of the disagreement set is unique and feasible under randomization for one-to-one matching. (ML: 0.83)
- The paper introduces an alternating path/cycle decomposition of the disagreement set for many-to-one matching with capacity constraints, which is used to develop an efficient procedure for constructing a collection of alternating paths and cycles that satisfies the conditions in Theorem 4. (ML: 0.80)
- The authors also draw on graph-theoretic techniques from the literature, such as finding augmenting paths and performing Euler-tour decompositions. (ML: 0.76)
- The paper builds on previous work by Imbens and Rubin (2015) and Athey and Imbens (2019), who introduced the concept of alternating path/cycle decomposition for one-to-one matching. (ML: 0.74)
Abstract
Matching mechanisms play a central role in operations management across diverse fields including education, healthcare, and online platforms. However, experimentally comparing a new matching algorithm against a status quo presents some fundamental challenges due to matching interference, where assigning a unit in one matching may preclude its assignment in the other. In this work, we take a design-based perspective to study the design of randomized experiments to compare two predetermined matching plans on a finite population, without imposing outcome or behavioral models. We introduce the notation of a disagreement set, which captures the difference between the two matching plans, and show that it admits a unique decomposition into disjoint alternating paths and cycles with useful structural properties. Based on these properties, we propose the Alternating Path Randomized Design, which sequentially randomizes along these paths and cycles to effectively manage interference. Within a minimax framework, we optimize the conditional randomization probability and show that, for long paths, the optimal choice converges to $\sqrt{2}-1$, minimizing worst-case variance. We establish the unbiasedness of the Horvitz-Thompson estimator and derive a finite-population Central Limit Theorem that accommodates complex and unstable path and cycle structures as the population grows. Furthermore, we extend the design to many-to-one matchings, where capacity constraints fundamentally alter the structure of the disagreement set. Using graph-theoretic tools, including finding augmenting paths and Euler-tour decomposition on an auxiliary unbalanced directed graph, we construct feasible alternating path and cycle decompositions that allow the design and inference results to carry over.
Matching mechanisms play a central role in operations management across diverse fields including education, healthcare, and online platforms. However, experimentally comparing a new matching algorithm against a status quo presents some fundamental challenges due to matching interference, where assigning a unit in one matching may preclude its assignment in the other. In this work, we take a design-based perspective to study the design of randomized experiments to compare two predetermined matching plans on a finite population, without imposing outcome or behavioral models. We introduce the notation of a disagreement set, which captures the difference between the two matching plans, and show that it admits a unique decomposition into disjoint alternating paths and cycles with useful structural properties. Based on these properties, we propose the Alternating Path Randomized Design, which sequentially randomizes along these paths and cycles to effectively manage interference. Within a minimax framework, we optimize the conditional randomization probability and show that, for long paths, the optimal choice converges to $\sqrt{2}-1$, minimizing worst-case variance. We establish the unbiasedness of the Horvitz-Thompson estimator and derive a finite-population Central Limit Theorem that accommodates complex and unstable path and cycle structures as the population grows. Furthermore, we extend the design to many-to-one matchings, where capacity constraints fundamentally alter the structure of the disagreement set. Using graph-theoretic tools, including finding augmenting paths and Euler-tour decomposition on an auxiliary unbalanced directed graph, we construct feasible alternating path and cycle decompositions that allow the design and inference results to carry over.
Why we are recommending this paper?
Due to your Interest in Design Patterns
The paper's focus on matching mechanisms is relevant to your interest in programming paradigms, particularly in the context of system design and algorithm development. Understanding how matching algorithms perform is crucial for effective programming.
Due to your Interest in Design Patterns
The paper's focus on matching mechanisms is relevant to your interest in programming paradigms, particularly in the context of system design and algorithm development. Understanding how matching algorithms perform is crucial for effective programming.
University of Copenhagen
AI Insights
- Language-specific neurons are concentrated in a small number of layers, most notably Layer 2 and the top layers, rather than being evenly distributed throughout the network. (ML: 0.97)
- ParsBERT encodes function-word categories in a more abstract and transferable manner across languages, whereas content-word representations are more language-specific and therefore less directly aligned across Persian and the test languages. (ML: 0.97)
- Case selectivity is concentrated in a small subset of layers rather than being evenly distributed across the network depth, with INS accounting for the largest share of case-selective neurons. (ML: 0.95)
- ParsBERT's ability to encode function-word categories in a more abstract manner across languages is notable, while content-word representations are more language-specific. (ML: 0.95)
- The analysis reveals that ParsBERT allocates a large number of highly specialized neurons to process Japanese tokens, indicating strong script-level and orthographic differences between Japanese and Persian. (ML: 0.95)
- LAPE: Layer-wise Analysis of Per-Category Embeddings UPOS: Universal Part-of-Speech tags The analysis provides insights into how ParsBERT represents language-specific information, including script-level and orthographic differences. (ML: 0.94)
- Gender selectivity is not uniformly distributed across depth, but concentrates in a small number of layers, especially early (Layer 2) and late (Layer 12), driven primarily by the NEUT category. (ML: 0.91)
- The LAPE results indicate that UPOS selectivity is present both very early and again in the upper part of the network, with weaker neuron-level selectivity in the intermediate layers. (ML: 0.87)
Abstract
We investigate structural traces of language contact in the intermediate representations of a monolingual language model. Focusing on Persian (Farsi) as a historically contact-rich language, we probe the representations of a Persian-trained model when exposed to languages with varying degrees and types of contact with Persian. Our methodology quantifies the amount of linguistic information encoded in intermediate representations and assesses how this information is distributed across model components for different morphosyntactic features. The results show that universal syntactic information is largely insensitive to historical contact, whereas morphological features such as Case and Gender are strongly shaped by language-specific structure, suggesting that contact effects in monolingual language models are selective and structurally constrained.
We investigate structural traces of language contact in the intermediate representations of a monolingual language model. Focusing on Persian (Farsi) as a historically contact-rich language, we probe the representations of a Persian-trained model when exposed to languages with varying degrees and types of contact with Persian. Our methodology quantifies the amount of linguistic information encoded in intermediate representations and assesses how this information is distributed across model components for different morphosyntactic features. The results show that universal syntactic information is largely insensitive to historical contact, whereas morphological features such as Case and Gender are strongly shaped by language-specific structure, suggesting that contact effects in monolingual language models are selective and structurally constrained.
Why we are recommending this paper?
Due to your Interest in Programming Language Design
This research explores the impact of language contact on language models, a potentially interesting area given your interest in programming language design and the influence of external factors on software development.
Due to your Interest in Programming Language Design
This research explores the impact of language contact on language models, a potentially interesting area given your interest in programming language design and the influence of external factors on software development.
Oklahoma Christian University
AI Insights
- L1 penalty: A regularization term that encourages sparse interaction schema and reduces overfitting. (ML: 0.95)
- MLP Baseline: A simple neural network architecture with multiple layers, which serves as a baseline model for comparison. (ML: 0.94)
- The paper presents a new model architecture called StructuralCFN, which is designed to handle tabular data and provides built-in interpretability. (ML: 0.92)
- TabNet: A library for tabular data processing that uses attention mechanisms to learn feature importance and weights. (ML: 0.90)
- StructuralCFN achieves state-of-the-art results on several benchmark datasets, including California Housing, Breast Cancer, Heart Disease, Wine Quality, Diabetes, and Ionosphere. (ML: 0.90)
- XGBoost: An implementation of gradient-boosted decision trees, which is used as a baseline model in the paper. (ML: 0.87)
- The StructuralCFN model achieves state-of-the-art results on several benchmark datasets and offers significant efficiency advantages over black-box ensembles. (ML: 0.84)
- The model's efficiency is attributed to its parallel-first architecture, which allows for constant-time execution of the dependency layer and aggregator on modern GPU/TPU hardware. (ML: 0.78)
- Future work will investigate hybrid forest-CFN architectures for high-entropy regimes, where discrete decision boundaries are more efficient to model. (ML: 0.76)
- StructuralCFN offers a significant scaling advantage over black-box ensembles in scientific manifolds where interactions are governed by a few dominant drivers. (ML: 0.69)
Abstract
Despite the ubiquity of tabular data in high-stakes domains, traditional deep learning architectures often struggle to match the performance of gradient-boosted decision trees while maintaining scientific interpretability. Standard neural networks typically treat features as independent entities, failing to exploit the inherent manifold structural dependencies that define tabular distributions. We propose Structural Compositional Function Networks (StructuralCFN), a novel architecture that imposes a Relation-Aware Inductive Bias via a differentiable structural prior. StructuralCFN explicitly models each feature as a mathematical composition of its counterparts through Differentiable Adaptive Gating, which automatically discovers the optimal activation physics (e.g., attention-style filtering vs. inhibitory polarity) for each relationship. Our framework enables Structured Knowledge Integration, allowing domain-specific relational priors to be injected directly into the architecture to guide discovery. We evaluate StructuralCFN across a rigorous 10-fold cross-validation suite on 18 benchmarks, demonstrating statistically significant improvements (p < 0.05) on scientific and clinical datasets (e.g., Blood Transfusion, Ozone, WDBC). Furthermore, StructuralCFN provides Intrinsic Symbolic Interpretability: it recovers the governing "laws" of the data manifold as human-readable mathematical expressions while maintaining a compact parameter footprint (300--2,500 parameters) that is over an order of magnitude (10x--20x) smaller than standard deep baselines.
Despite the ubiquity of tabular data in high-stakes domains, traditional deep learning architectures often struggle to match the performance of gradient-boosted decision trees while maintaining scientific interpretability. Standard neural networks typically treat features as independent entities, failing to exploit the inherent manifold structural dependencies that define tabular distributions. We propose Structural Compositional Function Networks (StructuralCFN), a novel architecture that imposes a Relation-Aware Inductive Bias via a differentiable structural prior. StructuralCFN explicitly models each feature as a mathematical composition of its counterparts through Differentiable Adaptive Gating, which automatically discovers the optimal activation physics (e.g., attention-style filtering vs. inhibitory polarity) for each relationship. Our framework enables Structured Knowledge Integration, allowing domain-specific relational priors to be injected directly into the architecture to guide discovery. We evaluate StructuralCFN across a rigorous 10-fold cross-validation suite on 18 benchmarks, demonstrating statistically significant improvements (p < 0.05) on scientific and clinical datasets (e.g., Blood Transfusion, Ozone, WDBC). Furthermore, StructuralCFN provides Intrinsic Symbolic Interpretability: it recovers the governing "laws" of the data manifold as human-readable mathematical expressions while maintaining a compact parameter footprint (300--2,500 parameters) that is over an order of magnitude (10x--20x) smaller than standard deep baselines.
Why we are recommending this paper?
Due to your Interest in Functional Programming
Due to your Interest in Functional Programming
University of Chinese Academy of Sciences, China
AI Insights
- NH-Rep: A method for learning implicit representations of CAD models using neural networks. (ML: 0.91)
- InstantNGP: A method for learning implicit representations of 3D objects using neural networks. (ML: 0.88)
- SharpNet is a new approach for learning sharp features in CAD models using neural networks. (ML: 0.87)
- SharpNet is a powerful tool for learning sharp features in CAD models, outperforming other state-of-the-art methods. (ML: 0.87)
- It uses a multi-resolution representation of the model and a novel loss function that encourages the network to produce sharp features. (ML: 0.86)
- SharpNet outperforms other state-of-the-art methods in reconstructing CAD models with sharp features, including NH-Rep and InstantNGP. (ML: 0.83)
- SDF: Signed distance function, a mathematical representation of the distance from a point to a surface. (ML: 0.78)
- The method can handle both open and closed sharp feature curves, unlike previous approaches which often fail on open curves. (ML: 0.77)
- It has the potential to revolutionize the field of computer-aided design by enabling more accurate and efficient reconstruction of complex CAD models. (ML: 0.77)
- CSG: Constructive solid geometry, a method for representing and manipulating 3D objects as combinations of simple shapes. (ML: 0.70)
Abstract
Multi-layer perceptrons (MLPs) are a standard tool for learning and function approximation, but they inherently yield outputs that are globally smooth. As a result, they struggle to represent functions that are continuous yet deliberately non-differentiable (i.e., with prescribed $C^0$ sharp features) without relying on ad hoc post-processing. We present SharpNet, a modified MLP architecture capable of encoding functions with user-defined sharp features by enriching the network with an auxiliary feature function, which is defined as the solution to a Poisson equation with jump Neumann boundary conditions. It is evaluated via an efficient local integral that is fully differentiable with respect to the feature locations, enabling our method to jointly optimize both the feature locations and the MLP parameters to recover the target functions/models. The $C^0$-continuity of SharpNet is precisely controllable, ensuring $C^0$-continuity at the feature locations and smoothness elsewhere. We validate SharpNet on 2D problems and 3D CAD model reconstruction, and compare it against several state-of-the-art baselines. In both types of tasks, SharpNet accurately recovers sharp edges and corners while maintaining smooth behavior away from those features, whereas existing methods tend to smooth out gradient discontinuities. Both qualitative and quantitative evaluations highlight the benefits of our approach.
Multi-layer perceptrons (MLPs) are a standard tool for learning and function approximation, but they inherently yield outputs that are globally smooth. As a result, they struggle to represent functions that are continuous yet deliberately non-differentiable (i.e., with prescribed $C^0$ sharp features) without relying on ad hoc post-processing. We present SharpNet, a modified MLP architecture capable of encoding functions with user-defined sharp features by enriching the network with an auxiliary feature function, which is defined as the solution to a Poisson equation with jump Neumann boundary conditions. It is evaluated via an efficient local integral that is fully differentiable with respect to the feature locations, enabling our method to jointly optimize both the feature locations and the MLP parameters to recover the target functions/models. The $C^0$-continuity of SharpNet is precisely controllable, ensuring $C^0$-continuity at the feature locations and smoothness elsewhere. We validate SharpNet on 2D problems and 3D CAD model reconstruction, and compare it against several state-of-the-art baselines. In both types of tasks, SharpNet accurately recovers sharp edges and corners while maintaining smooth behavior away from those features, whereas existing methods tend to smooth out gradient discontinuities. Both qualitative and quantitative evaluations highlight the benefits of our approach.
Why we are recommending this paper?
Due to your Interest in Functional Programming
Due to your Interest in Functional Programming
Interests not found
We did not find any papers that match the below interests.
Try other terms also consider if the content exists in arxiv.org.
- Programming Paradigms
You can edit or add more interests any time.
💬 Help Shape Our Pricing
We're exploring pricing options to make this project sustainable. Take 3 minutes to share what you'd be willing to pay (if anything). Your input guides our future investment.
Help us improve your experience!
This project is on its early stages your feedback can be pivotal on the future of the project.
Let us know what you think about this week's papers and suggestions!
Give Feedback