Code Generation & Program Synthesis

Cross-lingual type representations can be extracted from untyped code, revealing hidden structures in state-of-the-art code models that challenge our understanding of their internal workings.

LLMs can generate diverse resident personas that produce executable smart home interaction schedules, eliminating the need for intrusive real-world data collection.

Learning the generation order in multimodal tasks can boost performance by over 4%—a game changer for DLMs.

Procedural similarity can dramatically enhance repository-level code generation, achieving a 41.14% Pass@1 score that outstrips traditional methods.

AI-teacher collaboration outperforms imitation learning, boosting student performance by nearly 50% on challenging coding tasks.

CodeTracer can pinpoint the exact backdoor data responsible for unsafe code completions, even in the face of sophisticated attacks.

Unbounded loops in hybrid quantum programs can now be effectively analyzed for resource consumption and termination, filling a critical gap in quantum program verification.

Naive verification methods for PLCs can produce 44% false alarms due to unrealistic sensor models, but a new hardware-faithful approach eliminates these errors entirely.

XAlpha revolutionizes alpha discovery by turning it into a continuous learning process that adapts and evolves based on real-time feedback.

Strong execution in LLMs doesn't equate to effective educational control, as they struggle to lower cognitive demand despite being able to increase it.

Self-evolving LLM agents can slash latency by up to 62% while significantly boosting reliability in industrial applications.

Transforming LLM prototypes into auditable agents can ensure compliance and safety without sacrificing performance, achieving full utility in complex enterprise applications.

Aleena revolutionizes research software collaboration by ensuring that the rationale behind decisions is preserved across diverse communication channels.

LLM-generated skills fail to outperform basic task prompts in data science workflows, challenging the assumption that automated skill generation enhances AI performance.

Constrained decoding in diffusion models can boost accuracy by over 20% on complex tasks without significant latency penalties.

This AI assistant transforms how students grasp complex computational concepts by providing contextualized, example-driven learning without giving away answers.

SolSmith uncovered 25 hidden miscompilation bugs in the Solidity compiler, revealing critical vulnerabilities that could jeopardize smart contracts.

Personalization in AI-driven development can introduce significant biases, with age and gender influencing the very structure of generated code.

Achieving 99.94% of the theoretical maximum for entanglement certification reveals a new frontier in the efficiency of quantum software development.

Process features can triple the authorship attribution accuracy in educational programming contexts, revealing the limitations of relying solely on final code submissions.

SynapseFlow achieves over 4x higher branch coverage and uncovers critical bugs that other tools miss, revolutionizing fuzz harness generation.

TrajSpec transforms vague bug reports into actionable specifications, boosting APR success rates by over 18% through structured evidence gathering.

Existing benchmarks fail to reveal the true performance capabilities of LLMs, with only 6.11% showing significant speed advantages over traditional implementations.

Bug reports that work for humans can actually hinder AI agents, with localization cues being critical for repair success.

Overcoming the challenge of over-merging, this methodology achieves a 99% AUC in identity resolution while drastically reducing mega-cluster sizes in the World of Code dataset.

ATLAS automates the transition from high-level deep learning models to FPGA implementations, drastically reducing the manual effort required for custom hardware acceleration.

Optimization performance varies significantly by workload, challenging the notion that larger models are always superior in coding tasks.

Coding agents can now be evaluated on tasks that truly test their problem-solving skills, rather than their ability to recall previously seen solutions.

AI's role in code review is not a simple enhancement; it hinges on human expertise and the review process structure, revealing a complex interplay that challenges prevailing assumptions.

Trustworthy code generation can be achieved without post-generation adjustments, improving security and functionality simultaneously.

A staggering 65.8% energy reduction is possible with the right design strategy, but beware the "memory wall" that can obliterate those gains at scale.

fog enables a dramatic leap in motion recognition accuracy, allowing users to intuitively express complex emotions and movements in animations.

Coding agents struggle with native language tasks, achieving only 78.7% resolution in a benchmark designed to reflect real customer requests in Russian.

Knowledge Debt is a silent threat to developer expertise, but it can be mitigated by designing AI agents that actively promote incidental learning.

DebugTracker reveals that understanding the debugging process can significantly enhance educational assessments, moving beyond just final code quality to the intricacies of student reasoning and problem-solving.

Achieving a 24-bit watermark payload for code attribution without needing model access, this method outperforms existing techniques under multiple attack scenarios.

Automating cost function generation for steganography with LLMs can boost security and efficiency, achieving a 46.3% increase in execution speed.

Achieving 100% error detection accuracy in smart home configurations could revolutionize user experience and safety in automation systems.

Thoughtful feature curation reveals that structural code complexity is a far stronger predictor of deployment risk than traditional change volume metrics.

A new evaluation framework reveals that current assessments of LLM-powered agents often misrepresent their true capabilities in real-world software development.

A general LLM code agent can autonomously prove all targeted lemmas in software verification, achieving unprecedented coverage without expert intervention.

Grounding tests in a specification boosts LLM code correctness by 38 percentage points, revealing that content trumps quantity in test effectiveness.

Visual fidelity in web app generation doesn't guarantee functional interaction, as evidenced by a leading model scoring 7.5 on interaction inference while trailing others by over 5x.

Resolve rates mask critical insights about coding agent performance, but TraceProbe uncovers the hidden trajectory structures that explain why some runs succeed while others fail.

Clumping errors in author identity mapping can lead to a staggering misrepresentation of developer contributions, with previous maps inflating precision metrics by failing to account for conflated identities.

LLMs can accurately identify falsified software engineering definitions but paradoxically reject many correct ones, revealing a troubling bias in their understanding.

Reusing existing language models for software engineering texts significantly outperforms training new domain-specific models from scratch, challenging assumptions about domain adaptation strategies.

Modular task decomposition in AI-generated analyses boosts transparency and reliability, enabling smaller models to outperform larger counterparts.

RAG can reduce hallucinations in LLM-generated API code, but it risks introducing unnecessary parameters when endpoints are known.

Coding agents can generate observability artifacts, but they miss key diagnostic semantics, exposing fault signals for only 13.99% of failures.