Multimodal search agents answer visual questions by interleaving image understanding, web retrieval, tool use, and evidence synthesis. Strong systems exist, but in two expensive regimes: proprietary frontier models such as GPT-5 and Gemini, or large open vision-language backbones trained with substantial agentic data and reinforcement learning. We ask a different question: when released agent trajectories are distilled into much smaller backbones under a single-node budget, what is actually transferred? We study this with LiteSearch-VL, a low-compute recipe for Qwen3-VL-2B and Qwen3-VL-4B that uses only released OpenSearch-VL trajectories, parameter-efficient LoRA adapters, and synthetic step-level preferences: DPO on GPT-5-generated hard negatives targeting five local failure modes (premature answer, wrong tool, weak query, repeated query, ignored image). Across 12,400 GPT-5-judged rollouts on SimpleVQA, FVQA, LiveVQA, and VDR-Bench-testmini, the dominant effect is behavioral rather than a uniform accuracy lift: full-trajectory supervised fine-tuning transfers the agent contract, taking the 2B model from almost never emitting a usable answer (1,237/1,240 no_answer rollouts) to 28.4% macro Pass@1, matching or slightly exceeding the off-the-shelf 4B base (25.6%). Synthetic preference learning and compact tool distillation act as refinements rather than phase transitions (best 4B configuration: 30.8% macro Pass@1). Finally, a controlled VDR step-budget ablation shows that extra search turns convert abstentions into wrong_entity errors rather than correct answers, identifying answer verification, not search depth, as the next bottleneck for small multimodal agents.
Saeed Khaki, Nima Safaei, Kamal Ginotra· 0 citations
Extreme low-bit inference offers a route toward smaller models and constrained deployment. Ternary language models restrict weights to $\{-1,0,+1\}$, approaching the limit of $\log_2 3 \approx 1.585$ bits/weight. The practical question for a pretrained model is not simply whether weights can be quantised but which capabilities survive and whether it remains useful for adaptation. We explore this by converting Qwen3.5-0.8B (752M parameters) to ternary weights using 72.4M tokens of quantisation-aware training (QAT). The resulting model, Cloe, is evaluated across 29 benchmarks, representation diagnostics, and downstream fine-tuning. The evidence shows non-uniform degradation. A linear probe recovers 43.76% of MMLU answers from the full-precision teacher's representations but only 26.19% from Cloe (near chance), indicating specialist factual information is lost. However, Cloe retains measurable performance on ten tasks, averaging 77.1% of teacher performance. Crucially, fine-tuning raises Cloe to 89.8% on SST-2 (95.6% of the matched teacher) and reaches 79.4% teacher retention on XSum. We attribute degradation to a combination of quantisation-induced information loss and incomplete recovery due to the limited QAT budget. We also highlight an evaluation pitfall: standard answer-letter scoring failed (Cloe emitted"A"on 98.6% of MMLU questions), necessitating continuation scoring. Ultimately, ternary conversion is unsuitable as a drop-in general replacement yet remains valuable as a compact substrate for task-specific models.
The remarkable performance of multimodal large language models (MLLMs) comes at the cost of substantial computational overhead, posing significant challenges to real-time deployment and cost effectiveness. Existing model routing approaches either decide from coarse request-level features alone or spend one or several extra language model passes to inspect the generated response, leaving the token-level uncertainty signals that emerge during generation unused. To address these limitations, we propose Pro-Router, a token-aware progressive model routing method with adaptive edge-cloud collaboration for efficient multimodal LLM inference. Pro-Router employs a two-stage progressive decision mechanism. First, a lightweight prompt pre-scorer module performs rapid pre-screening before token generation begins, guiding apparently simple requests to small models. Second, a token-aware verifier reads the sampling probability distribution of each token the small model generates, estimating the model's confidence in its own output to determine, per request, whether the answer ships or escalates to the cloud-based high-precision model. Furthermore, we design an adaptive edge-cloud serving pipeline that sizes every dispatch to each device's measured service rate, so both the edge and the cloud tiers stay fully utilized without manual parameter tuning and are not impacted by the network latency. Extensive experiments on multiple multimodal benchmark datasets and models demonstrate the effectiveness of Pro-Router. Compared to other methods, it achieves the highest routing accuracy and improves routing speed by more than 10x. Its serving pipeline also reaches more than 75% higher end-to-end throughput than the existing model routing pipeline. Our code is available at https://github.com/xinyuangui2/pro-router.
Xin-Yuan Gui, Shao-Wen Wang, Sheng Sun et al.· 0 citations
The results raise concerns that LLM-assisted evaluation may under-select proposals that human reviewers identify as highly novel, potentially reflecting the statistical logic of next-token prediction trained on past scientific outputs.
Diogo Machado· Scientometrics· 0 citations
Reach audiences
Advertise in front of researchers, engineers, and readers.
A constrained formulation in which the tools that carry class discriminative signal form a mandatory core placed outside the policy’s reach, while the remaining choice minimizes the measured invocation cost, which is subjected to an explicit relative accuracy floor.
The rapid advancement of large language models (LLMs) offers transformative potential for secondary physics education. However, theoretical and practical frameworks for responsible, pedagogically sound LLM integration remain underdeveloped. This paper proposes a structured, theory-grounded framework for embedding LLMs into secondary physics instruction, informed by constructivism, cognitive load theory, and the TPACK model. We identify three core application modalities: teacher-facing instructional design tools, guarded student-facing inquiry tools, and assessment-augmentation tools. We analyze domain-specific opportunities (targeting misconceptions in mechanics and electromagnetism) and risks (cognitive offloading, epistemic opacity, algorithmic bias). Drawing on empirical evidence from a small-scale single-class pilot feasibility study rather than systematic full empirical verification
,
we formalize six pedagogical principles: progressive autonomy, epistemic transparency, cognitive load optimization, disciplinary fidelity, ethical scaffolding, and human–AI complementarity. A one-group pretest–posttest pilot exploration was implemented with 36 secondary school students (ages 14–16) learning Newton’s Third Law solely to examine the preliminary practicability of the proposed framework instead of validating its universal effectiveness. Data analyses included paired-samples t-tests, effect size calculation, compliance coding, and descriptive survey statistics. Results showed significantly improved conceptual understanding within this limited pilot sample (
p
< 0.001, d = 1.32, large effect), high in-class observed adherence to AI use guardrails (91.7%), and generally positive learner perceptions among the participating cohort. This pilot-tested operable framework clarifies the evolving role of physics teachers as orchestrators of human–AI collaboration and offers actionable preliminary guidance for educators, instructional designers, and policymakers.
The next big jump will not be driven by larger parameters, but smaller, more visible and dynamic systems that can be trusted by security analysts, as well as a disconnect between lab and reality.
Hai-Yang Wang, Yuejin Zhang· Frontiers in Computing and I...· 0 citations
Three-dimensional printing is currently closer to routine clinical use than vascular bioprinting, particularly in anatomical modeling, treatment planning, procedural simulation, education, and patient communication.
Agnieszka Sawina, Sara Rakotoarison, Kaja Hanys et al.· Archiv Euromedica· 0 citations
Clinical deployment of artificial intelligence-generated radiology reports requires explicit entity-relation validation, structured accuracy assessment, and radiologist oversight to protect patient safety, and these findings reinforce a broader principle.
Unknown authors· Frontiers in Digital Health· 0 citations
The Concept Tree Learner is introduced, an incremental and interpretable symbolic framework that induces logical concepts over binary strings from minimal labeled data and generalizes substantially better to unseen strings than both a classical entropy-based decision tree and the RIPPER rule learner.
Muhammed Tekin Ertekin, Burkay Genç· Applied Sciences· 0 citations
This study designs, prototypes, and evaluates Servi.AI, a personality-aware multi-agent intelligent decision support system for enterprise strategy, and finds that selective rather than default deployment is supported.
Xu Zhou, Zhong-Yi Jiang· Applied System Innovation· 0 citations
The writing skills of second-grade students at SD Negeri 4 Made Lamongan remain low, as shown by approximately 60% of students who have not reached the Minimum Completeness Criteria, particularly in composing simple sentences and developing vocabulary. Indonesian language learning that is still dominated by lecture-based methods without varied media is one of the causes of students' low engagement. This study aims to produce a “Petualangan Kata” educational game media based on Word Search Maker that is valid, practical, and effective in improving the writing skills of second-grade elementary students. This study used the Research and Development (R&D) method with the ADDIE development model (Analysis, Design, Development, Implementation, Evaluation) and a Control Group Pretest-Posttest experimental design. The research subjects were 40 second-grade students of SD Negeri 4 Made, consisting of class II A as the experimental group and class II B as the control group. The results showed that the media was categorized as highly valid, with average scores of 3.76 from material experts, 3.84 from media experts, and 3.82 from design experts out of a maximum score of 4.00. The practicality of the media was also in the very good category, with student response scores of 3.75 in the small-group trial and 3.81 in the large-group trial. The t-test result obtained a significance value of 0.000 (< 0.05), indicating a significant difference between the experimental and control classes, while the N-Gain Score test showed a 73.66% improvement in the experimental class, categorized as high. This study concludes that the “Petualangan Kata” educational game media based on Word Search Maker is valid, practical, and effective for improving the writing skills of second-grade elementary school students.
ABSTRAK
Keterampilan menulis siswa kelas II Sekolah Dasar Negeri 4 Made Lamongan masih tergolong rendah, terlihat dari sekitar 60% siswa yang belum mencapai Kriteria Ketuntasan Minimal, terutama pada kemampuan menyusun kalimat sederhana dan mengembangkan kosakata. Pembelajaran Bahasa Indonesia yang masih didominasi metode ceramah tanpa variasi media menjadi salah satu penyebab rendahnya keterlibatan siswa. Penelitian ini bertujuan menghasilkan media game edukasi “Petualangan Kata” berbasis Word Search Maker yang valid, praktis, dan efektif untuk meningkatkan keterampilan menulis siswa kelas II SD. Penelitian menggunakan metode Research and Development (R&D) dengan model pengembangan ADDIE (Analysis, Design, Development, Implementation, Evaluation) serta desain eksperimen Control Group Pretest-Posttest. Subjek penelitian adalah 40 siswa kelas II SD Negeri 4 Made, terdiri atas kelas II A sebagai kelompok eksperimen dan kelas II B sebagai kelompok kontrol. Hasil penelitian menunjukkan media dinyatakan sangat valid dengan skor rata-rata ahli materi 3,76, ahli media 3,84, dan ahli desain 3,82 dari skor maksimal 4,00. Tingkat kepraktisan media juga berada pada kategori sangat baik, dengan skor respons siswa 3,75 pada uji coba kelompok kecil dan 3,81 pada uji coba kelompok besar. Hasil uji-t memperoleh nilai signifikansi 0,000 (< 0,05) yang menunjukkan perbedaan signifikan antara kelas eksperimen dan kelas kontrol, sementara uji N-Gain Score menunjukkan peningkatan sebesar 73,66% pada kelas eksperimen dengan kategori tinggi. Simpulan penelitian ini adalah media game edukasi “Petualangan Kata” berbasis Word Search Maker valid, praktis, dan efektif digunakan untuk meningkatkan keterampilan menulis siswa kelas II Sekolah Dasar.
What if pathology foundation models could do more with less? GigaPath-Flash and GigaTIME-Flash cut computational demands while maintaining strong performance, opening the door to larger studies and broader exploration. The post GigaPath-Flash and GigaTIME-Flash: Toward population-scale discovery with efficient pathology foundation models appeared first on Microsoft Research.
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
