{"id":1185241,"date":"2026-05-08T00:00:00","date_gmt":"2026-05-08T07:00:00","guid":{"rendered":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/?post_type=msr-research-item&#038;p=1185241"},"modified":"2026-09-01T13:59:25","modified_gmt":"2026-09-01T20:59:25","slug":"table-first-guarded-reasoning-outperforming-larger-mllms-in-chart-question-answering","status":"publish","type":"msr-research-item","link":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/publication\/table-first-guarded-reasoning-outperforming-larger-mllms-in-chart-question-answering\/","title":{"rendered":"Table-First Guarded Reasoning: Outperforming Larger MLLMs in Chart Question Answering"},"content":{"rendered":"\n\n\n<p class=\"wp-block-paragraph\">Multimodal Large Language Models (MLLMs) have revolutionized visual understanding but frequently suffer from &#8220;visual hallucination&#8221; when extracting precise numerical values from charts. Existing solutions often mitigate this by scaling parameters to massive sizes (e.g., 70B+ models), creating significant computational barriers for resource-constrained environments. To address this, we introduce Table-First Guarded Reasoning (TFGR), a resource-efficient pipeline that decouples visual data extraction from logical reasoning. By integrating a dedicated chart-to-table extraction module (Chart2Table) with Zero-Shot Chain-of-Thought (CoT) prompting, we provide the model with structured tabular context to &#8220;guard&#8221; against perception errors. We evaluate our approach on the ChartQA benchmark using the Qwen2.5-VL-7B-Instruct model. Our pipeline achieves 93.96% accuracy, significantly outperforming the 88.00% baseline. Notably, our 8.87B-parameter architecture (when accounting for Chart2Table) surpasses the State-of-the-Art Qwen2-VL-72B (88.20%) [1] on the OpenVLM Leaderboard, demonstrating that architectural innovation can substitute for raw parameter scale. We further validate our findings through an extensive analysis of explorative experiments in fine-tuning, architectural alternatives, and complementary tools.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Multimodal Large Language Models (MLLMs) have revolutionized visual understanding but frequently suffer from &#8220;visual hallucination&#8221; when extracting precise numerical values from charts. Existing solutions often mitigate this by scaling parameters to massive sizes (e.g., 70B+ models), creating significant computational barriers for resource-constrained environments. To address this, we introduce Table-First Guarded Reasoning (TFGR), a resource-efficient pipeline [&hellip;]<\/p>\n","protected":false},"featured_media":0,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":true,"_classifai_error":"","msr-author-ordering":[{"type":"text","value":"A. Aziz","user_id":0},{"type":"text","value":"M. Abbas","user_id":0},{"type":"text","value":"Kirolous Fouty","user_id":0},{"type":"text","value":"Shaza Ali","user_id":0},{"type":"text","value":"Magde Mohamed","user_id":0},{"type":"text","value":"T. Kassab","user_id":0},{"type":"user_nicename","value":"Ahmed Tawfik","user_id":"31140"},{"type":"text","value":"Hossam Sharara","user_id":0},{"type":"text","value":"Cherif R. Salama","user_id":0}],"msr_publishername":"","msr_publisher_other":"","msr_booktitle":"","msr_chapter":"","msr_edition":"Conference on Algebraic Informatics","msr_editors":"","msr_how_published":"","msr_isbn":"","msr_issue":"","msr_journal":"2026 IEEE Conference on Artificial Intelligence (CAI)","msr_number":"","msr_organization":"IEEE","msr_pages_string":"1599\u20131604","msr_page_range_start":"1599","msr_page_range_end":"1604","msr_series":"","msr_volume":"","msr_copyright":"\u00a9 2026 IEEE","msr_conference_name":"2026 IEEE Conference on Artificial Intelligence (CAI)","msr_doi":"10.1109\/CAI68641.2026.11536412","msr_arxiv_id":"","msr_mag_id":"","msr_other_authors":"","msr_other_contributors":"","msr_speaker":"","msr_award":"","msr_affiliation":"","msr_institution":"","msr_host":"","msr_version":"","msr_duration":"","msr_release_tracker_id":"","msr_highlight_type":"","msr_date_display_format":"","msr_main_download_label":"","msr_external_link_label":"","msr_doi_label":"","msr_published_date":"2026-05-08","msr_startdate":"","msr_presentation_date":"","msr_highlight_text":"","msr_notes":"","msr_longbiography":"","msr_publicationurl":"","msr_external_url":"","msr_secondary_video_url":"","msr_conference_url":"","msr_journal_url":"","msr_year":2026,"msr_month":5,"msr_day":8,"msr_microsoftintellectualproperty":false,"msr_pub_id":"249a3a955eb377b151b9d8f505021684ac4e1eaa","msr_publication_uploader":[{"type":"file","title":"Table-First_Guarded_Reasoning_Outperforming_Larger_MLLMs_in_Chart_Question_Answering.pdf","label_id":243132,"id":1185244,"viewUrl":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Table-First_Guarded_Reasoning_Outperforming_Larger_MLLMs_in_Chart_Question_Answering.pdf"},{"type":"url","title":"https:\/\/www.computer.org\/csdl\/proceedings-article\/cai\/2026\/11536412\/2gXMqXL9u7e","label_id":0,"id":false,"viewUrl":false}],"msr_related_uploader":[],"msr_original_fields_of_study":[],"msr_s2_paper_id":"","msr_s2_pdf_url":"","msr_citation_count_updated":"","msr_citation_count":0,"msr_influential_citations":0,"msr_reference_count":0,"msr_s2_open_access":false,"msr_s2_author_ids":[],"msr_pub_ids":[{"provider":"s2","id":"249a3a955eb377b151b9d8f505021684ac4e1eaa"},{"provider":"doi","id":"10.1109\/CAI68641.2026.11536412"},{"provider":"corpusid","id":"288870892"}],"msr_hide_image_in_river":0,"footnotes":""},"msr-research-highlight":[],"research-area":[13556],"msr-publication-type":[193716],"msr-publisher":[],"msr-publication-cta":[],"msr-focus-area":[243450,243210,243600,243477,243594],"msr-locale":[268875],"msr-post-option":[],"msr-field-of-study":[249796,270561,248338],"msr-conference":[],"msr-journal":[270559],"msr-impact-theme":[264846],"msr-pillar":[],"class_list":["post-1185241","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-artificial-intelligence","msr-focus-area-image-processing","msr-focus-area-machine-learning","msr-focus-area-machine-reading-comprehension","msr-focus-area-model-training","msr-focus-area-natural-language-processing","msr-locale-en_us","msr-field-of-study-ai","msr-field-of-study-chart-understanding","msr-field-of-study-question-answering"],"msr_publishername":"","msr_edition":"Conference on Algebraic Informatics","msr_affiliation":"","msr_published_date":"2026-05-08","msr_host":"","msr_duration":"","msr_version":"","msr_speaker":"","msr_other_contributors":"","msr_booktitle":"","msr_pages_string":"1599\u20131604","msr_chapter":"","msr_isbn":"","msr_journal":"2026 IEEE Conference on Artificial Intelligence (CAI)","msr_volume":"","msr_number":"","msr_editors":"","msr_series":"","msr_issue":"","msr_organization":"IEEE","msr_how_published":"","msr_notes":"","msr_highlight_text":"","msr_release_tracker_id":"","msr_original_fields_of_study":"","msr_download_urls":"","msr_external_url":"","msr_secondary_video_url":"","msr_longbiography":"","msr_microsoftintellectualproperty":0,"msr_main_download":"","msr_publicationurl":"","msr_doi":"10.1109\/CAI68641.2026.11536412","msr_publication_uploader":[{"type":"file","title":"Table-First_Guarded_Reasoning_Outperforming_Larger_MLLMs_in_Chart_Question_Answering.pdf","label_id":243132,"id":1185244,"viewUrl":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Table-First_Guarded_Reasoning_Outperforming_Larger_MLLMs_in_Chart_Question_Answering.pdf"},{"type":"url","title":"https:\/\/www.computer.org\/csdl\/proceedings-article\/cai\/2026\/11536412\/2gXMqXL9u7e","label_id":0,"id":false,"viewUrl":false}],"msr_related_uploader":[],"msr_citation_count":0,"msr_citation_count_updated":"","msr_s2_paper_id":"","msr_influential_citations":0,"msr_reference_count":0,"msr_arxiv_id":"","msr_s2_author_ids":[],"msr_s2_open_access":false,"msr_s2_pdf_url":null,"msr_attachments":[{"id":1185244,"url":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Table-First_Guarded_Reasoning_Outperforming_Larger_MLLMs_in_Chart_Question_Answering.pdf"}],"msr-author-ordering":[{"type":"text","value":"A. Aziz","user_id":0,"rest_url":false},{"type":"text","value":"M. Abbas","user_id":0,"rest_url":false},{"type":"text","value":"Kirolous Fouty","user_id":0,"rest_url":false},{"type":"text","value":"Shaza Ali","user_id":0,"rest_url":false},{"type":"text","value":"Magde Mohamed","user_id":0,"rest_url":false},{"type":"text","value":"T. Kassab","user_id":0,"rest_url":false},{"type":"user_nicename","value":"Ahmed Tawfik","user_id":31140,"rest_url":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Ahmed Tawfik"},{"type":"text","value":"Hossam Sharara","user_id":0,"rest_url":false},{"type":"text","value":"Cherif R. Salama","user_id":0,"rest_url":false}],"msr_impact_theme":["Computing foundations"],"msr_research_lab":[],"msr_event":[],"msr_group":[],"msr_project":[],"publication":[],"video":[],"msr-tool":[],"msr_publication_type":"inproceedings","related_content":[],"_links":{"self":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1185241","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item"}],"about":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-research-item"}],"version-history":[{"count":10,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1185241\/revisions"}],"predecessor-version":[{"id":1185252,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1185241\/revisions\/1185252"}],"wp:attachment":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1185241"}],"wp:term":[{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=1185241"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1185241"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=1185241"},{"taxonomy":"msr-publisher","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publisher?post=1185241"},{"taxonomy":"msr-publication-cta","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-cta?post=1185241"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=1185241"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1185241"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1185241"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=1185241"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=1185241"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=1185241"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1185241"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=1185241"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}