Search for: All records

Abstract This paper presents a new approach to improve static program analysis using Large Language Models (LLMs). The approachinterleavescalls to the static analyzer and queries to the LLM. The query to the LLM is constructed based on intermediate results from the static analysis, and subsequent static analysis uses the results from the LLM query. We apply our approach to the problem oferror-specification inference: given systems code written in C, infer the set of values that each function can return on error. Such error specifications aid in program understanding and can be used to find error-handling bugs. We implemented our approach by incorporating LLMs into EESI, the state-of-the-art static analysis for error-specification inference. Compared to EESI, our approach achieves higher recall (from an average of 52.55% to 77.83%) and higher F1-score (from an average of 0.612 to 0.804) while maintaining precision (from an average of 86.67% to 85.12%) on real-world benchmarks such as Apache HTTPD and MbedTLS. We also conducted experiments to understand the sources of imprecision in our LLM-assisted analysis as well as the impact of LLM nondeterminism on the analysis results. 

As the COVID-19 pandemic progresses, widespread community transmission of SARS-CoV-2 has ushered in a volatile era of viral immune evasion rather than the much-heralded stability of “endemicity” or “herd immunity.” At this point, an array of viral strains has rendered essentially all monoclonal antibody therapeutics obsolete and strongly undermined the impact of vaccinal immunity on SARS-CoV-2 transmission. In this work, we demonstrate that antibody escape resulting in evasion of pre-existing immunity is highly evolutionarily favored and likely to cause waves of short-term transmission. In the long-term, invading strains that induce weak cross-immunity against pre-existing strains may co-circulate with those pre-existing strains. This would result in the formation of serotypes that increase disease burden, complicate SARS-CoV-2 control, and raise the potential for increases in viral virulence. Less durable immunity does not drive positive selection as a trait, but such strains may transmit at high levels if they establish. Overall, our results draw attention to the importance of inter-strain cross-immunity as a driver of transmission trends and the importance of early immune evasion data to predict the trajectory of the pandemic. 

Deep Neural Networks (DNNs) have grown in popularity over the past decade and are now being used in safety-critical domains such as aircraft collision avoidance. This has motivated a large number of techniques for finding unsafe behavior in DNNs. In contrast, this paper tackles the problem of correcting a DNN once unsafe behavior is found. We introduce the provable repair problem, which is the problem of repairing a network N to construct a new network N′ that satisfies a given specification. If the safety specification is over a finite set of points, our Provable Point Repair algorithm can find a provably minimal repair satisfying the specification, regardless of the activation functions used. For safety specifications addressing convex polytopes containing infinitely many points, our Provable Polytope Repair algorithm can find a provably minimal repair satisfying the specification for DNNs using piecewise-linear activation functions. The key insight behind both of these algorithms is the introduction of a Decoupled DNN architecture, which allows us to reduce provable repair to a linear programming problem. Our experimental results demonstrate the efficiency and effectiveness of our Provable Repair algorithms on a variety of challenging tasks. 

Abstract Pellet-based extrusion deposition of carbon fiber-reinforced composites at high material deposition rates has recently gained much attention due to its applications in large-scale additive manufacturing. The mechanical and physical properties of large-volume components largely depend on their reinforcing fiber length. However, very few studies have been done thus far to have a direct comparison of additively fabricated composites reinforced with different carbon fiber lengths. In this study, a new additive manufacturing (AM) approach to fabricate long fiber-reinforced polymer (LFRP) was first proposed. A pellet-based extrusion deposition method was implemented, which directly used thermoplastic pellets and continuous fiber tows as feedstock materials. Discontinuous long carbon fibers, with an average fiber length of 20.1 mm, were successfully incorporated into printed LFRP samples. The printed LFRP samples were compared with short fiber-reinforced polymer (SFRP) and continuous fiber-reinforced polymer (CFRP) counterparts through mechanical tests and microstructural analyses. The carbon fiber dispersion, distribution of carbon fiber length and orientation, and fiber wetting were studied. As expected, a steady increase in flexural strength was observed with increasing fiber length. The carbon fibers were highly oriented along the printing direction. A more uniformly distributed discontinuous fiber reinforcement was found within printed SFRP and LFRP samples. Due to decreased fiber impregnation time and lowered impregnation rate, the printed CFRP samples showed a lower degree of impregnation and worse fiber wetting conditions. The feasibility of the proposed AM methods was further demonstrated by fabricating large-volume components with complex geometries.