[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85168-en":3,"doc-seo-85168-105":29,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},85168,962075114765,"Quinn","https://ap-avatar.wpscdn.com/davatar_a8503ba1806abce46bf441b54a3ca4cd",8,"Research & Report","A Symbolic Neural CPU for Quantization-Simulated Writeback and Interpretable Program Execution","Neural networks can learn algorithmic input-output mappings, yet their executors are hard to trust because state transitions are typically opaque. This work presents a trace-supervised symbolic neural CPU with a factorized learned execution architecture: recurrent control, an explicit operation router over a differentiable arithmetic-logic unit bank, destination-masked register writeback, trajectory supervision, and matched fixed-point replay. The trace includes selected operations, source/destination registers, register trajectories, memory signals, and writeback semantics. On a 16-wide benchmark, exact reproduction holds for non-quantized execution and the symbolic operation path is preserved under simulated eight-bit quantization.","arXiv :2607 . 1002 1v 1 [ cs .AI] 10 Jul 2026  \nA SYMBOLIC NEURAL CPU FOR QUANTIZATION-SIMULATED WRITEBACK AND INTERPRETABLE PROGRAM EXECUTION  \nJose Luis Lima de Jesus Silva* 1,2  \n1Federal University of Bahia, Department of Geophysics, Salvador, BA 40170-115, Brazil  \n2 Grupo de Estudos e Aplicação de Inteligência Artificial em Geofísica (GAIA), Federal University of Bahia, Salvador,  \nBA 40170-115, Brazil  \nABSTRACT  \nNeural networks can learn algorithmic input-output mappings, yet learned executors remain difficult to trust because their internal state transitions are usually hidden. We introduce a trace-supervised symbolic neural CPU, a factorized learned execution architecture that combines recurrent control, an explicit operation router over a fixed differentiable arithmetic-logic unit bank, destination-masked register writeback, complete trajectory supervision and matched fixed-point replay. The resulting execution process is auditable through the selected operation, source and destination registers, register trajectory, memory signals and writeback semantics. On the principal 16-wide benchmark, thenon-quantized executor reproduces reference execution exactly, while the eight-bit quantizationsimulated executor preserves the symbolic operation path through programs of 1000 instructions.  \nResidual numerical drift disappears against a matched fixed-point replay, identifying a mismatch between continuous and low-precision reference semantics rather than execution failure. Architecture comparisons span recurrent, Transformer, temporal-convolution, temporal graph-inspired and statespace controllers. Ablations show that operation-gate supervision is necessary for an inspectable execution path, while hidden-opcode memory-pressure tasks expose the current limits of delayed state use and temporal binding. We further extend the interface with ValueMemory, hybrid adaptive leaky integrate-and-fire controllers, candidate-constrained symbolic control trained through behavior cloning and actor-critic reinforcement learning, and an RV32I base-integer semantic bridge. Together, these results establish a trace-verifiable framework for interpretable, low-precision and controllable neural execution.  \n1 Introduction  \nModern computers are not judged only by their final outputs. They are auditable transition systems where instructions select operations, operations act on addressable state, and the resulting trajectory can be inspected step by step. This separation between control, arithmetic, and memory is central to the stored-program view of computation, where variables can be manipulated by changing the addresses from which values are read and to which values are written [1–3] . Modern neural networks have followed a different route. They learn continuous input-output maps from data and are usually evaluated by task-level accuracy, even when the task itself requires an algorithmic procedure. The difficulty is therefore not simply whether a neural model can approximate the input-output behavior of a program, but whether it can expose an execution process sufficiently faithful to be audited as computation.  \nThis issue has shaped a long line of work at the boundary between learning, memory and reasoning. Recurrent neural networks (RNNs) introduced trainable hidden states for sequence processing, and long short-term memory (LSTM) networks improved the ability of recurrent models to retain information over extended time intervals [4–6] . The sequence-to-sequence model, attention mechanisms, and pointer-like architectures further showed that neural networks can learn flexible forms of alignment, retrieval, and structured sequential manipulation [7–9] . These new developments made neural systems increasingly capable of acting over temporally extended inputs, but the stored  \n∗ [Corresponding author. Email](Corresponding author. Email mailto:jseluis.silva@gmail.com)[ mailto:jseluis.silva@gmail.com](Corresponding author. Email mailto:jseluis.si","cbCaihTVeSnnGym2","https://ap.wps.com/l/cbCaihTVeSnnGym2","pdf",8293343,1,63,"English","en",105,"# Introduction\n## Auditable transition systems vs. neural input-output maps\n## Memory-augmented neural computation\n## Neural algorithmic reasoning and intermediate supervision","[{\"question\":\"Why are learned neural executors difficult to trust in algorithmic tasks?\",\"answer\":\"Their internal state transitions are usually hidden, making it hard to audit how computation proceeds beyond final task-level accuracy.\"},{\"question\":\"What new architecture is proposed to enable interpretable, auditable execution?\",\"answer\":\"A trace-supervised symbolic neural CPU that uses recurrent control, an explicit operation router over a fixed differentiable ALU bank, destination-masked register writeback, and complete trajectory supervision with matched fixed-point replay.\"},{\"question\":\"How does quantization simulation affect execution correctness and interpretability?\",\"answer\":\"On the principal 16-wide benchmark, the non-quantized executor reproduces reference execution exactly, while the eight-bit quantization-simulated executor preserves the symbolic operation path through programs of 1000 instructions.\"}]",1784201511,159,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"a-symbolic-neural-cpu-for-quantization-simulated-writeback-and-interpretable-program-execution","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/a-symbolic-neural-cpu-for-quantization-simulated-writeback-and-interpretable-program-execution/85168/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"Why are learned neural executors difficult to trust in algorithmic tasks?","Question",{"text":75,"@type":76},"Their internal state transitions are usually hidden, making it hard to audit how computation proceeds beyond final task-level accuracy.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What new architecture is proposed to enable interpretable, auditable execution?",{"text":80,"@type":76},"A trace-supervised symbolic neural CPU that uses recurrent control, an explicit operation router over a fixed differentiable ALU bank, destination-masked register writeback, and complete trajectory supervision with matched fixed-point replay.",{"name":82,"@type":73,"acceptedAnswer":83},"How does quantization simulation affect execution correctness and interpretability?",{"text":84,"@type":76},"On the principal 16-wide benchmark, the non-quantized executor reproduces reference execution exactly, while the eight-bit quantization-simulated executor preserves the symbolic operation path through programs of 1000 instructions.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":45,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":45,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":45,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]