讓前沿 AI 落地真實產業場景Turning frontier AI into capabilities industry can actually use. Turning frontier AI into capabilities industry can actually use.
深港人工智能應用創新中心，專注邊緣 AI、端側推理與國產算力（信創），依託深港雙城資源，打通「前沿研究 — 產業落地」，把方案真正部署到政企現場。 A Hong Kong–based innovation centre that pairs Hong Kong's international research and capital with Shenzhen's engineering and scale — specialising in edge AI, on-device inference and domestic compute, and bringing frontier research to real government and enterprise sites.
深港人工智能應用創新中心，是一個專注人工智能前沿技術轉化、與產業共創落地場景的平台。The Centre is a platform dedicated to translating frontier AI research into real industrial applications — co-created with the industries that put them to work.
我們把最新的 AI 研究成果，對接到真實的政企場景中：依託深圳與香港的雙城資源，讓技術不止於論文，而能落到產線、櫃枱與終端設備上。We connect the latest AI research to real government and enterprise scenarios. Drawing on the combined resources of Hong Kong and Shenzhen, we take technology beyond the paper — onto production lines, service counters and edge devices.
從研究、適配到部署，我們關注的始終是同一件事 —— 把前沿能力，變成客戶當下用得上的結果。From research and adaptation to deployment, we focus on one thing throughout — turning frontier capability into results clients can use today.
深圳的工程與製造、香港的研究與國際化，在中心交匯成一條從前沿到落地的完整鏈路。Hong Kong's research and global reach, Shenzhen's engineering and manufacturing — converging into a single path from frontier research to deployment.
Where ideas become hardware
Where research meets the world
中心位於深港之間 —— 把香港的研究、資本與合規，接上深圳的算力、製造與場景，讓前沿研究與產業落地，在同一條鏈路上完成。Positioned between the two cities, the Centre connects Hong Kong's research, capital and compliance with Shenzhen's compute, manufacturing and scenarios — so frontier research and industrial deployment happen along a single chain.
圍繞「端側智能」展開，從算力底座到產業落地，形成可交付的完整能力。Built around on-device intelligence — from the compute foundation to industrial deployment, delivered end to end.
在終端設備上完成 AI 推理：低延遲、數據不出本地、離線也能用，兼顧效能與隱私合規。AI inference that runs on the device itself: low latency, data that never leaves, and offline operation — balancing performance with privacy and compliance.
適配國產 AI 芯片與信創軟硬件棧，合規、可控、可交付，貼合政企採購與自主可控要求。Adapted to domestic AI chips and the Xinchuang stack — compliant, controllable and deliverable, aligned with public-sector procurement and self-reliance requirements.
連接兩地的研究、資本與場景，把方案真正部署到產業現場，讓深港的優勢在同一個項目裡匯流。Connecting research, capital and scenarios across both cities, we deploy solutions on site — bringing the strengths of Hong Kong and Shenzhen together in a single project.
中心不只做技術，更串起生態各方 —— 讓前沿研究、國產算力、金融科技與政企場景，在深港之間協同落地。Beyond technology, the Centre links the players — bringing frontier research, domestic compute, fintech and real-world scenarios together across Shenzhen and Hong Kong.
尤其聚焦金融科技與政企場景 —— 在對數據安全、合規與穩定性要求最高的領域，提供可靠、可交付的端側 AI 方案。With a focus on fintech and the public sector — delivering dependable, deployable on-device AI where data security, compliance and reliability matter most.
金融、政務、公共安全的人工智能，必須跑在境內、跑在國產算力上、跑在客戶的合規邊界之內。這些約束在工程上是麻煩，在研究上卻是一片長期空白 —— 大量方法學只在公開數據集上驗證過，沒有人在真實的受監管部署環境裡檢驗過它們的前提是否還成立。我們把這些場景開放出來，尋找願意在真實約束下做研究的高校實驗室。AI for finance, government and public security has to run inside the mainland, on domestic compute, within the client's compliance boundary. Engineering-wise these constraints are an inconvenience; research-wise they are a long-standing blind spot — a great deal of methodology has only ever been validated on public benchmarks, never where its assumptions actually have to hold. We are opening these settings up, and looking for university labs willing to do research under real constraints.
中心不授學位、不佔招生指標、不參與錄取。學籍、導師關係、學術評價與畢業標準，全部留在高校一側。研究方向由高校導師決定，中心不干預學術結論。The Centre does not award degrees, does not hold admission quotas, and takes no part in admissions decisions. Enrolment, supervision, academic assessment and graduation standards all remain with the university. Research direction is set by the academic supervisor; we do not interfere with academic conclusions.
目前正在與內地及香港的高校實驗室推進首批共建，以單實驗室、單課題、單學生為起步單元。We are currently establishing our first partnerships with labs in the mainland and Hong Kong, starting with one lab, one topic, one student.
視覺 token 壓縮是端側部署的必選項。已有工作表明，壓縮會放大而非削弱對抗脆弱性 —— 它更像一個失真濃縮器而非降噪器。但這些威脅模型尚未在真實部署環境中檢驗過：實際的攻擊面、算力預算與輸入分佈，都與實驗室設定不同。Visual token compression is not optional at the edge. Recent work shows compression amplifies rather than dampens adversarial vulnerability — behaving as a distortion concentrator rather than a denoiser. Those threat models have yet to meet a real deployment, where the attack surface, the compute budget and the input distribution all differ from the lab setting.
數據不能出機構，模型卻需要持續更新。通信效率與隱私保護之間的取捨，在真實的機構間約束下究竟是什麼形狀。Data cannot leave the institution; the model still has to keep learning. What the communication-efficiency and privacy trade-off actually looks like under real cross-institutional constraints.
被遺忘權與監管審計都要求「可證明的數據刪除」。什麼樣的刪除是可證明的、審計方認什麼 —— 學術定義與監管實踐之間目前有一段距離。The right to be forgotten and regulatory audit both demand provable deletion. Which deletions are provable, and which ones auditors accept — there is a gap here between the academic definition and regulatory practice.
端側可部署、跨法域適用的小模型。真正的難點不在模型，在於「什麼算個人信息」本身隨法域而變。Small models, deployable at the edge, usable across jurisdictions. The hard part is not the model; it is that what counts as personal information changes with the jurisdiction.
量化、剪枝與編譯優化在改變模型效率的同時，也在改變它的安全屬性。這部分幾乎沒有系統性的測量。Quantisation, pruning and compiler optimisation change a model's efficiency and its security properties at the same time. The second half is barely measured.
課題以雙方共同確定為準，以上是可以立即啟動的方向。Topics are agreed jointly. The above are directions we can start on now.
研究在境內、在客戶的合規邊界內進行，數據不出域。中心以供應商身份服務受監管客戶，不代表客戶，亦不承諾任何數據的直接提供；每一課題可觸及的數據範圍，以客戶單獨授權與合規審查結論為準。Research happens inside the jurisdiction and inside the client's compliance boundary; the data stays put. The Centre serves regulated clients as a vendor — we do not represent them, and we do not undertake to supply data. What any given project can touch is determined by that client's own authorisation and compliance review.
方法與結論經脫敏審查後可正常發表。審查範圍僅限客戶身份、業務數據與部署細節，窗口 15 個工作日，無異議即視為通過；中心不以審查為由干預學術結論。Methods and conclusions are publishable once desensitisation review is complete. Review covers client identity, business data and deployment specifics only, with a 15-working-day window; no response means cleared. Review is never a lever over academic conclusions.
學生的學術成果與學位論文權利歸學生與高校。中心不主張對通用方法學的獨佔；僅當技術與中心既有產品或特定場景直接結合、形成可專利成果時，雙方就共有與實施許可另行約定。Academic outputs and thesis rights belong to the student and the university. The Centre claims no exclusivity over general methodology; only where the work combines directly with an existing product or a specific setting to form a patentable result do we agree separate terms on joint ownership and licensing.
學生畢業與發表，不受商業條款掣肘。這是前提，不是讓步。A student's graduation and publication are not subject to commercial terms. That is the premise, not a concession.
合作、洽談或諮詢，歡迎隨時來信，我們會盡快回覆。For partnerships, projects or enquiries, write to us anytime — we'll reply promptly.
lxy@shengangai.com