Harvard University
AI Insights
- The concept of intrinsic motivation is central to understanding human behavior and decision-making. [3]
- Intrinsic motivation can be influenced by various factors, including curiosity, creativity, and a sense of accomplishment. [3]
- Prediction error refers to the difference between what is expected and what actually happens. [3]
- Intrinsic motivation can be influenced by various factors, including personality traits, cognitive abilities, and environmental conditions. [3]
- Intrinsic motivation: The drive to engage in an activity for its own sake, rather than for external rewards or pressures. [3]
- Prediction error: The difference between what is expected and what actually happens. [3]
- The concept of intrinsic motivation is complex and multifaceted. [3]
- It can be influenced by various factors, including personality traits, cognitive abilities, and environmental conditions. [3]
- Intrinsic motivation plays a crucial role in human behavior and decision-making. [3]
- It drives individuals to engage in activities for their own sake, rather than for external rewards or pressures. [3]
- The idea of using prediction error as a driving force for exploration is explored in several studies. [2]
- Intrinsic motivation refers to the drive to engage in an activity for its own sake, rather than for external rewards or pressures. [1]
Abstract
Where do objective functions come from? How do we select what goals to pursue? Human intelligence is adept at synthesizing new objective functions on the fly. How does this work, and can we endow artificial systems with the same ability? This paper proposes an approach to answering these questions, starting with the concept of a subjective function, a higher-order objective function that is endogenous to the agent (i.e., defined with respect to the agent's features, rather than an external task). Expected prediction error is studied as a concrete example of a subjective function. This proposal has many connections to ideas in psychology, neuroscience, and machine learning.
Where do objective functions come from? How do we select what goals to pursue? Human intelligence is adept at synthesizing new objective functions on the fly. How does this work, and can we endow artificial systems with the same ability? This paper proposes an approach to answering these questions, starting with the concept of a subjective function, a higher-order objective function that is endogenous to the agent (i.e., defined with respect to the agent's features, rather than an external task). Expected prediction error is studied as a concrete example of a subjective function. This proposal has many connections to ideas in psychology, neuroscience, and machine learning.
Why we are recommending this paper?
Due to your Interest in: Functional Programming
This paper explores the concept of objective functions, a core element in programming language design and optimization – aligning directly with your interest in programming paradigms. It delves into how systems can learn and adapt goals, a fascinating area relevant to functional programming.
Due to your Interest in: Functional Programming
This paper explores the concept of objective functions, a core element in programming language design and optimization – aligning directly with your interest in programming paradigms. It delves into how systems can learn and adapt goals, a fascinating area relevant to functional programming.
Indiana University Bloom
Abstract
Automated floor plan generation lies at the intersection of combinatorial search, geometric constraint satisfaction, and functional design requirements -- a confluence that has historically resisted a unified computational treatment. While recent deep learning approaches have improved the state of the art, they often struggle to capture architectural reasoning: the precedence of topological relationships over geometric instantiation, the propagation of functional constraints through adjacency networks, and the emergence of circulation patterns from local connectivity decisions. To address these fundamental challenges, this paper introduces GFLAN, a generative framework that restructures floor plan synthesis through explicit factorization into topological planning and geometric realization. Given a single exterior boundary and a front-door location, our approach departs from direct pixel-to-pixel or wall-tracing generation in favor of a principled two-stage decomposition. Stage A employs a specialized convolutional architecture with dual encoders -- separating invariant spatial context from evolving layout state -- to sequentially allocate room centroids within the building envelope via discrete probability maps over feasible placements. Stage B constructs a heterogeneous graph linking room nodes to boundary vertices, then applies a Transformer-augmented graph neural network (GNN) that jointly regresses room boundaries.
Automated floor plan generation lies at the intersection of combinatorial search, geometric constraint satisfaction, and functional design requirements -- a confluence that has historically resisted a unified computational treatment. While recent deep learning approaches have improved the state of the art, they often struggle to capture architectural reasoning: the precedence of topological relationships over geometric instantiation, the propagation of functional constraints through adjacency networks, and the emergence of circulation patterns from local connectivity decisions. To address these fundamental challenges, this paper introduces GFLAN, a generative framework that restructures floor plan synthesis through explicit factorization into topological planning and geometric realization. Given a single exterior boundary and a front-door location, our approach departs from direct pixel-to-pixel or wall-tracing generation in favor of a principled two-stage decomposition. Stage A employs a specialized convolutional architecture with dual encoders -- separating invariant spatial context from evolving layout state -- to sequentially allocate room centroids within the building envelope via discrete probability maps over feasible placements. Stage B constructs a heterogeneous graph linking room nodes to boundary vertices, then applies a Transformer-augmented graph neural network (GNN) that jointly regresses room boundaries.
Why we are recommending this paper?
Due to your Interest in: Functional Programming
Given your interest in programming language design, this paper's focus on automated layout generation for functional programs is highly relevant. The work explores a computational approach to designing functional systems, a key area of interest.
Due to your Interest in: Functional Programming
Given your interest in programming language design, this paper's focus on automated layout generation for functional programs is highly relevant. The work explores a computational approach to designing functional systems, a key area of interest.
Beihang University
AI Insights
- Reinforcement Learning from Human Feedback (RLHF): A technique used to train LLMs by providing them with feedback from humans, such as ratings or corrections. [3]
- The paper discusses the concept of large language models (LLMs) and their applications in various domains. [2]
Abstract
Code large language models (Code LLMs) are powerful but costly to train, with scaling laws predicting performance from model size, data, and compute. However, different programming languages (PLs) have varying impacts during pre-training that significantly affect base model performance, leading to inaccurate performance prediction. Besides, existing works focus on language-agnostic settings, neglecting the inherently multilingual nature of modern software development. Therefore, it is first necessary to investigate the scaling laws of different PLs, and then consider their mutual influences to arrive at the final multilingual scaling law. In this paper, we present the first systematic exploration of scaling laws for multilingual code pre-training, conducting over 1000+ experiments (Equivalent to 336,000+ H800 hours) across multiple PLs, model sizes (0.2B to 14B parameters), and dataset sizes (1T tokens). We establish comprehensive scaling laws for code LLMs across multiple PLs, revealing that interpreted languages (e.g., Python) benefit more from increased model size and data than compiled languages (e.g., Rust). The study demonstrates that multilingual pre-training provides synergistic benefits, particularly between syntactically similar PLs. Further, the pre-training strategy of the parallel pairing (concatenating code snippets with their translations) significantly enhances cross-lingual abilities with favorable scaling properties. Finally, a proportion-dependent multilingual scaling law is proposed to optimally allocate training tokens by prioritizing high-utility PLs (e.g., Python), balancing high-synergy pairs (e.g., JavaScript-TypeScript), and reducing allocation to fast-saturating languages (Rust), achieving superior average performance across all PLs compared to uniform distribution under the same compute budget.
Code large language models (Code LLMs) are powerful but costly to train, with scaling laws predicting performance from model size, data, and compute. However, different programming languages (PLs) have varying impacts during pre-training that significantly affect base model performance, leading to inaccurate performance prediction. Besides, existing works focus on language-agnostic settings, neglecting the inherently multilingual nature of modern software development. Therefore, it is first necessary to investigate the scaling laws of different PLs, and then consider their mutual influences to arrive at the final multilingual scaling law. In this paper, we present the first systematic exploration of scaling laws for multilingual code pre-training, conducting over 1000+ experiments (Equivalent to 336,000+ H800 hours) across multiple PLs, model sizes (0.2B to 14B parameters), and dataset sizes (1T tokens). We establish comprehensive scaling laws for code LLMs across multiple PLs, revealing that interpreted languages (e.g., Python) benefit more from increased model size and data than compiled languages (e.g., Rust). The study demonstrates that multilingual pre-training provides synergistic benefits, particularly between syntactically similar PLs. Further, the pre-training strategy of the parallel pairing (concatenating code snippets with their translations) significantly enhances cross-lingual abilities with favorable scaling properties. Finally, a proportion-dependent multilingual scaling law is proposed to optimally allocate training tokens by prioritizing high-utility PLs (e.g., Python), balancing high-synergy pairs (e.g., JavaScript-TypeScript), and reducing allocation to fast-saturating languages (Rust), achieving superior average performance across all PLs compared to uniform distribution under the same compute budget.
Why we are recommending this paper?
Due to your Interest in: Programming Language Design
This paper directly addresses the impact of programming languages on model performance, which is a crucial aspect of programming language design and evaluation. Understanding how different languages scale is essential for informed design choices.
Due to your Interest in: Programming Language Design
This paper directly addresses the impact of programming languages on model performance, which is a crucial aspect of programming language design and evaluation. Understanding how different languages scale is essential for informed design choices.
ACM
AI Insights
- The study evaluates the code reasoning abilities of large language models (LLMs) on real-world problems. [3]
- RE2-Bench is a comprehensive benchmark that includes problems with different difficulty levels, including simple and complex synthetic programs. [3]
- False negatives in input prediction are plausible, as multiple inputs may result in the same output. [3]
- Automating the detection of false negatives in input prediction faces non-trivial challenges, such as initializing classes with complex attributes before calling methods. [2]
Abstract
Code reasoning tasks are becoming prevalent in large language model (LLM) assessments. Existing benchmarks involve simple programs, failing to represent real-world complexities such as inter- or intra-procedural dependencies, core or third-party API calls, highly nested constructs, and non-primitive complex types. Evaluating LLMs under such a simplistic setting poses a significant threat to assumptions about their generalizability in practice. To enable a more realistic evaluation of code reasoning, this paper proposes RE2-Bench, a benchmark of 1,101 reasoning problems, including 195 drawn from mature real-world projects. RE2-Bench leverages static and dynamic program analysis to automatically serialize and deserialize compound, complex, and custom types in real-world code, going far beyond the primitive-only settings used in prior work. A key feature of RE2-Bench is categorizing each reasoning problem as Easy or Hard via a principled majority-vote mechanism over nine interpretable code complexity metrics, resulting in two well-separated and semantically meaningful difficulty categories suitable for precise calibration of LLM reasoning ability. A comprehensive evaluation of six general-purpose and reasoning-oriented LLMs on two widely used code reasoning tasks -- input prediction and output prediction -- using RE2-Bench reveals a significant performance drop from Easy to Hard problems (51.50\% for input prediction and 42.15\% for output prediction), confirming that prior evaluations substantially overestimate the reasoning capabilities of LLMs.
Code reasoning tasks are becoming prevalent in large language model (LLM) assessments. Existing benchmarks involve simple programs, failing to represent real-world complexities such as inter- or intra-procedural dependencies, core or third-party API calls, highly nested constructs, and non-primitive complex types. Evaluating LLMs under such a simplistic setting poses a significant threat to assumptions about their generalizability in practice. To enable a more realistic evaluation of code reasoning, this paper proposes RE2-Bench, a benchmark of 1,101 reasoning problems, including 195 drawn from mature real-world projects. RE2-Bench leverages static and dynamic program analysis to automatically serialize and deserialize compound, complex, and custom types in real-world code, going far beyond the primitive-only settings used in prior work. A key feature of RE2-Bench is categorizing each reasoning problem as Easy or Hard via a principled majority-vote mechanism over nine interpretable code complexity metrics, resulting in two well-separated and semantically meaningful difficulty categories suitable for precise calibration of LLM reasoning ability. A comprehensive evaluation of six general-purpose and reasoning-oriented LLMs on two widely used code reasoning tasks -- input prediction and output prediction -- using RE2-Bench reveals a significant performance drop from Easy to Hard problems (51.50\% for input prediction and 42.15\% for output prediction), confirming that prior evaluations substantially overestimate the reasoning capabilities of LLMs.
Why we are recommending this paper?
Due to your Interest in: Programming Language Design
With your interest in programming paradigms, this paper’s focus on assessing code reasoning abilities in LLMs is a strong match. It investigates complex code scenarios, aligning with your interest in understanding how systems solve programming problems.
Due to your Interest in: Programming Language Design
With your interest in programming paradigms, this paper’s focus on assessing code reasoning abilities in LLMs is a strong match. It investigates complex code scenarios, aligning with your interest in understanding how systems solve programming problems.
Alfrd Rnnyi Institute
AI Insights
- The paper is about a mathematical concept called the Turan number. [3]
- The authors are looking at specific types of graphs, like nested and separated matchings, and they're trying to figure out how many edges can be in these graphs without having those patterns. [3]
- The paper discusses the Turan number of ordered graphs, which is a measure of the maximum number of edges in an ordered graph that does not contain certain patterns. [2]
Abstract
A {\it vertex-ordered} graph is a graph equipped with a linear ordering of its vertices. A pair of independent edges in an ordered graph can exhibit one of the following three patterns: separated, nested or crossing. We say a pair of independent edges is non-separated if it is either crossing or nested. Non-nested and non-crossing pairs are defined analogously. We are interested in the following Turán-type problems: for each of the aforementioned six patterns, determine the maximum number of edges of an $n$-vertex ordered graph that does not contain a $k$-matching such that every pair of edges exhibit the fixed pattern. Exact answers have already been obtained for four of the six cases. The main objective of this paper is to investigate the two remaining open cases, namely non-separated and non-nested matchings. We determine the exact maximum number of edges of an $n$-vertex ordered graph that does not contain a non-separated $k$-matching, which has the form $\frac{3}{2}(k-1)n+Θ(k^2)$. For the non-nested case, we show the maximum number of edges lies between $(k-1)n$ and $(k-1)n+\binom{k-1}{2}$. We also determine the exact maximum number of edges of an $n$-vertex ordered graph that does not contain an alternating path of given length. We discuss some related problems and raise several conjectures. Furthermore, our results and conjectures yield consequences to certain Ramsey-type problems for non-nested matchings and alternating paths.
A {\it vertex-ordered} graph is a graph equipped with a linear ordering of its vertices. A pair of independent edges in an ordered graph can exhibit one of the following three patterns: separated, nested or crossing. We say a pair of independent edges is non-separated if it is either crossing or nested. Non-nested and non-crossing pairs are defined analogously. We are interested in the following Turán-type problems: for each of the aforementioned six patterns, determine the maximum number of edges of an $n$-vertex ordered graph that does not contain a $k$-matching such that every pair of edges exhibit the fixed pattern. Exact answers have already been obtained for four of the six cases. The main objective of this paper is to investigate the two remaining open cases, namely non-separated and non-nested matchings. We determine the exact maximum number of edges of an $n$-vertex ordered graph that does not contain a non-separated $k$-matching, which has the form $\frac{3}{2}(k-1)n+Θ(k^2)$. For the non-nested case, we show the maximum number of edges lies between $(k-1)n$ and $(k-1)n+\binom{k-1}{2}$. We also determine the exact maximum number of edges of an $n$-vertex ordered graph that does not contain an alternating path of given length. We discuss some related problems and raise several conjectures. Furthermore, our results and conjectures yield consequences to certain Ramsey-type problems for non-nested matchings and alternating paths.
Why we are recommending this paper?
Due to your Interest in: Design Patterns
This paper's exploration of graph patterns and ordered vertices offers a potentially interesting perspective on algorithm design and data structures, aligning with your broader interest in programming paradigms. The focus on pattern recognition is a valuable skill in programming.
Due to your Interest in: Design Patterns
This paper's exploration of graph patterns and ordered vertices offers a potentially interesting perspective on algorithm design and data structures, aligning with your broader interest in programming paradigms. The focus on pattern recognition is a valuable skill in programming.
Bielefeld University
Abstract
We introduce graph pattern-based association rules (GPARs) for directed labeled multigraphs such as RDF graphs. GPARs support both generative tasks, where a graph is extended, and evaluative tasks, where the plausibility of a graph is assessed. The framework goes beyond related formalisms such as graph functional dependencies, graph entity dependencies, relational association rules, graph association rules, multi-relation and path association rules, and Horn rules. Given a collection of graphs, we evaluate graph patterns under no-repeated-anything semantics, which allows the topology of a graph to be taken into account more effectively. We define a probability space and derive confidence, lift, leverage, and conviction in a probabilistic setting. We further analyze how these metrics relate to their classical itemset-based counterparts and identify conditions under which their characteristic properties are preserved.
We introduce graph pattern-based association rules (GPARs) for directed labeled multigraphs such as RDF graphs. GPARs support both generative tasks, where a graph is extended, and evaluative tasks, where the plausibility of a graph is assessed. The framework goes beyond related formalisms such as graph functional dependencies, graph entity dependencies, relational association rules, graph association rules, multi-relation and path association rules, and Horn rules. Given a collection of graphs, we evaluate graph patterns under no-repeated-anything semantics, which allows the topology of a graph to be taken into account more effectively. We define a probability space and derive confidence, lift, leverage, and conviction in a probabilistic setting. We further analyze how these metrics relate to their classical itemset-based counterparts and identify conditions under which their characteristic properties are preserved.
Why we are recommending this paper?
Due to your Interest in: Design Patterns
Due to your Interest in: Design Patterns
IMT Atlantique
Abstract
This document, intended for computer science teachers, describes a case study that puts into practice a questioning of ethical, societal and environmental issues when designing or implementing a decision support system. This study is based on a very popular application, namely road navigation software that informs users of real-time traffic conditions and suggests routes between a starting point and a destination, taking these conditions into account (such as Waze). The approach proposes to intertwine technical considerations (optimal path algorithms, data needed for location, etc.) with a broader view of the ethical, environmental and societal issues raised by the tools studied. Based on the authors' experience conducting sessions with students over several years, this document discusses the context of such a study, suggests teaching resources for implementing it, describes ways to structure discussions, and shares scenarios in different teaching contexts.
This document, intended for computer science teachers, describes a case study that puts into practice a questioning of ethical, societal and environmental issues when designing or implementing a decision support system. This study is based on a very popular application, namely road navigation software that informs users of real-time traffic conditions and suggests routes between a starting point and a destination, taking these conditions into account (such as Waze). The approach proposes to intertwine technical considerations (optimal path algorithms, data needed for location, etc.) with a broader view of the ethical, environmental and societal issues raised by the tools studied. Based on the authors' experience conducting sessions with students over several years, this document discusses the context of such a study, suggests teaching resources for implementing it, describes ways to structure discussions, and shares scenarios in different teaching contexts.
AI Insights
- Data collection raises issues related to privacy, possible uses, quality, availability, etc. [3]
- The author suggests exploring technical alternatives that do not collect private data. [3]
- The author emphasizes that data collection is not imposed by the functionality of the tool, but is a choice made by its designer. [3]
- The article discusses the potential issues with navigation tools like Waze, including environmental and societal impacts. [2]
Why we are recommending this paper?
Due to your Interest in: Programming Paradigms
Due to your Interest in: Programming Paradigms
North Carolina State Unv
Abstract
Block-Based Programming (BBP) platforms, such as Snap!, have become increasingly prominent in K-12 computer science education due to their ability to simplify programming concepts and foster computational thinking from an early age. While these platforms engage students through visual and gamified interfaces, teachers often face challenges in using them effectively and finding all the necessary features for classroom management. To address these challenges, we introduce SnapClass, a classroom management system integrated within the Snap! programming environment. SnapClass was iteratively developed drawing on established research about the pedagogical and logistical challenges teachers encounter in computing classrooms. Specifically, SnapClass allows educators to create and customize block-based coding assignments based on student skill levels, implement rubric-based auto-grading, and access student code history and recovery features. It also supports monitoring student engagement and idle time, and includes a help dashboard with a raise hand feature to assist students in real time. This paper describes the design and key features of SnapClass those are developed and those are under progress.
Block-Based Programming (BBP) platforms, such as Snap!, have become increasingly prominent in K-12 computer science education due to their ability to simplify programming concepts and foster computational thinking from an early age. While these platforms engage students through visual and gamified interfaces, teachers often face challenges in using them effectively and finding all the necessary features for classroom management. To address these challenges, we introduce SnapClass, a classroom management system integrated within the Snap! programming environment. SnapClass was iteratively developed drawing on established research about the pedagogical and logistical challenges teachers encounter in computing classrooms. Specifically, SnapClass allows educators to create and customize block-based coding assignments based on student skill levels, implement rubric-based auto-grading, and access student code history and recovery features. It also supports monitoring student engagement and idle time, and includes a help dashboard with a raise hand feature to assist students in real time. This paper describes the design and key features of SnapClass those are developed and those are under progress.
AI Insights
- SnapClass includes an AI-based assistance for auto-grading, which allows teachers to create rubrics from scratch or generate one with AI. [3]
- The system also features a chatbot called SnapBot, which empowers students to access just-in-time support, especially helpful in large classrooms or asynchronous learning environments. [3]
- Future work will include empirical studies with more teachers and students, expanding AI capabilities to support formative feedback, and refining chatbot personalization for varied learning needs. [3]
- AI-based assistance: the use of artificial intelligence to aid in tasks such as auto-grading or providing feedback. [3]
- The use of AI-based assistance can help reduce teacher workload while maintaining pedagogical flexibility. [3]
- Empirical studies are needed to further evaluate the effectiveness of SnapClass in supporting problem-solving skills in programming classrooms. [3]
- Block-based programming: a visual approach to programming that uses blocks instead of text-based code. [2]
- The paper presents SnapClass, a system that integrates AI with block-based programming environments to provide scalable, real-time support for both instruction and assessment. [1]
Why we are recommending this paper?
Due to your Interest in: Object Oriented Programming
Due to your Interest in: Object Oriented Programming
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