- π Model Overview (λͺ¨λΈ κ°μ)
- π‘οΈ Defense & Performance Benchmarks (μ€μ λ°©μ΄ μ§ν)
- π Dataset & Threat Vector Distribution (νμ΅ λ°μ΄ν° ꡬμ±)
- β‘ Inference Speed & Runtime Benchmark (μΆλ‘ λ²€μΉλ§ν¬)
- π¦ Export Formats & Deployment Artifacts (λ°°ν¬ νμΌ κ΅¬μ±)
- π License & Commercial Terms (λΌμ΄μ μ€ λ° μμ
μ λμ
μλ΄)
- β οΈ Notes & Disclaimer (κΈ°μ κ³ μ§ λ° μλ΄)
- πΌ Opportunities & Contact (μ±μ© μ μ λ° ννΈλμ)
BAAR4-guard Ultra (In-Line Multi-Era AI Threat Engine)
π Official Website: https://baar.uk/model/
π¦ Model Hub: aixk/BAAR4-guard
π License: Non-Commercial Only (Contact for Commercial Pricing)
π Model Overview (λͺ¨λΈ κ°μ)
BAAR4-guard Ultra is an ultra-fast, in-line edge AI threat detection and firewall engine trained across historical and modern traffic benchmarks spanning 2017 to 2026 (CIC-IDS2017, CICIoT2023, CICDDoS2019, and 2024~2026 Synthetic AI Vectors). Hardened with Multi-Stage PGD (Projected Gradient Descent) Adversarial Training, the model provides zero-trust packet/session classification with extreme resilience against adversarial payload perturbations and evasive botnets, delivering sub-millisecond inference on edge and WASM runtimes.
BAAR4-guard Ultraλ 2017~2026 μ μ£ΌκΈ° μ€μ λ€νΈμν¬ νΈλν½(CIC-IDS2017, CICIoT2023, CICDDoS2019 λ° 2024~2026 μ΅μ AI 곡격 ν©μ± 벑ν°)μ ν΅ν© νμ΅ν μ΄κ²½λΒ·μ΄μ μ§μ° μΈλΌμΈ μΈκ³΅μ§λ₯ 보μ μμ§μ λλ€. νΉν **λ€λ¨κ³ PGD μ λμ νλ ¨(Adversarial Training)**μ μ μ©νμ¬ ν¨ν· λ³μ‘° λ° AI κΈ°λ° μ°ν 곡격(Adversarial Evasion)μ λν΄ μλμ μΈ λ°©μ΄ λ³΅μλ ₯μ λ°ννλ©°, μ€ν(FPR)μ μ΅μννλ©΄μ 0.8ms μμ€μ μΈλΌμΈ κ²μ¦ μλλ₯Ό 보μ₯ν©λλ€.
π‘οΈ Defense & Performance Benchmarks (μ€μ λ°©μ΄ μ§ν)
Validation Setup: PGD Lv.4 Adversarial Perturbation Test | Hardware: PyTorch CUDA & Edge Runtime Export
| Class / Category | Classification Type | Detection Rate (μ°¨λ¨μ¨ / ν΅κ³Όμ¨) | Error Metric (μ€λ₯μ¨) | Primary Threat Scenarios |
|---|---|---|---|---|
| Class 0: μ μ νΈλν½ (Normal Human) | Benign Flow | 99.84% ν΅κ³Όμ¨ | FPR: 0.16% |
μΌλ° μ 무 νΈλν½, μ μ IoT ν΅μ , μΈκ°λ HTTP/S νλ‘μ° |
| Class 1: AI μ μ°° / λ΄λ· (Recon/Botnet) | Threat Flow | 100.00% μ°¨λ¨μ¨ | FNR: 0.00% |
μμ¨ ν¬νΈμ€μΊ, μ·¨μ½μ μΈν 리μ μ€ μ€μΊλ, C2 λΉμ½ ν΅μ |
| Class 2: μΉ κ³΅κ²© / DoS (Web Attack & DoS) | Threat Flow | 100.00% μ°¨λ¨μ¨ | FNR: 0.00% |
SYN Flood, ICMP Flood, DrDoS (UDP/NetBIOS/Portmap), Web Exploit |
| Class 3: ν¬λ¦¬λ΄μ 곡격 (Credential Brute-force) | Threat Flow | 99.70% ~ 100.00% μ°¨λ¨μ¨ | FNR: <= 0.30% |
κ³μ νμ·¨ μλ, λΆμ° λΈλ£¨νΈν¬μ€, Credential Stuffing |
β‘ μ’ ν© λ°©μ΄ λ° μ λμ κ²¬κ³ μ± (Adversarial Robustness)
- μ€μ μ’
ν© κ³΅κ²© μ°¨λ¨μ¨ (Attack Block Rate): 99.97% ~ 100.00% (λ―Ένλ₯ FNR:
0.00% ~ 0.03%) - Lv.4 ν¨ν· λ³μ‘° AI λ°©μ΄λ ₯ (PGD Robustness): 99.11% ~ 99.58%
- μ μ νΈλν½ ν΅κ³Όμ¨ (Benign Pass Rate): 99.84% (μ€νλ₯ FPR:
0.16%) - μ΅μ’ κ²μ¦ λ³΅ν© μ μ (Overall Defense Score): 298.96 / 300.00
π Dataset & Threat Vector Distribution (νμ΅ λ°μ΄ν° ꡬμ±)
μ 체 νμ΅ μ μ aixk/bg-1μ ν¨ν€μ§λμ΄ λκΈ°νλμμΌλ©°, μ΄ 31,333건μ μμ λ μκ³μ΄ μλμ°/ν¨ν· νΌμ²λ‘ ꡬμ±λμμ΅λλ€:
π― [κ· ν λΉ
λ°μ΄ν°μ
μμ±] μ΄ 31,333κ° μ΅μ μκ³μ΄ μλμ°
β μ μ νλ‘μ° (Class 0) : 12,521건 (CIC-IDS2017 Tuesday + CICIoT2023 Benign + ν©μ± λ°μ΄ν°)
β AI μμ¨ μ μ°° (Class 1) : 2,321건 (PortScan + 2024~2026 AI Recon Vectors)
β DoS / μΉ κ³΅κ²© (Class 2): 15,000건 (SYN/ICMP Flood, DrDoS UDP/Portmap/NetBIOS, Web Attacks)
β ν¬λ¦¬λ΄μ
곡격 (Class 3) : 1,491건 (Thursday Web Attack Brute-force + ν©μ± Stuffing)
- CIC-IDS2017 Benchmark:
Tuesday-WorkingHours(μ μ 5,312건)Thursday-WorkingHours-Morning-WebAttacks(μΉ μΉ¨ν¬/λ¬΄μ°¨λ³ λμ 1,806건)Friday-WorkingHours-Afternoon-PortScan(λ€νΈμν¬ μ μ°° 4,213건)
- CICIoT2023 & CICDDoS2019 Large-Scale Vectors:
BenignTraffic,DDoS-ICMP_Flood,DDoS-SYN_Flood(CICIoT2023: μ΄ 8,856건)Syn,DrDoS_UDP,Portmap,DrDoS_NetBIOS(CICDDoS2019: μ΄ 10,162건)
- 2024~2026 μ΅μ μν μ£Όμ
κΈ° (Modern Injected Vectors):
- κ° ν΄λμ€λ³ 4,000κ±΄μ© μ΄ 16,000건μ μ°¨μΈλ λ€λ³λ λ³μ‘° ν¨ν΄ ν©μ± μ£Όμ
β‘ Inference Speed & Runtime Benchmark (μΆλ‘ λ²€μΉλ§ν¬)
2,000ν μ°μ λ¨μΌ ν¨ν·/μλμ° νλ³ λ²€μΉλ§ν¬ κ²°κ³Ό:
| Runtime Engine | Single Inference Latency | Model Binary Size | Runtime Characteristics |
|---|---|---|---|
| ONNX Runtime (CPU/WASM) | 0.882 ms | 2.53 MiB |
ν¬λ‘μ€ νλ«νΌ, μλν¬μΈνΈ Wasm ꡬλ μ΅μ ν |
| λΉμμΆ .baar3 Direct Engine | 1.788 ms | 1.24 MiB |
Zero-Copy Direct-mmap λ°μ΄λ리, μ΅μ λ©λͺ¨λ¦¬ ννλ¦°νΈ |
μ΄μ μ§μ° μΈλΌμΈ μ μ©: λ¨μΌ μΈμ λΆμ 1ms λ―Έλ§μ μ²λ¦¬ μ±λ₯μ μ μ§νμ¬ λ€νΈμν¬ λ³λͺ© νμ μλ ν¬λͺ ν μΈλΌμΈ L7/L4 λ°©νλ²½ ν΅ν©μ΄ κ°λ₯ν©λλ€.
π¦ Export Formats & Deployment Artifacts (λ°°ν¬ νμΌ κ΅¬μ±)
model.safetensors: κ³ μ λ‘λ© λ° λ³΄μ κ²μ¦μ ν΅κ³Όν μμ κ°μ€μΉ ν μ νμΌmodel.onnx: WebAssembly, C++, Go λ± μμ§ λ°νμ νΈν κ³ μ ONNX κ·Έλν (2.53 MiB)model_guard_ultra_fp16.baar3: Zero-Copy Direct-mmap κΈ°λ° FP16 μ΄κ²½λ λ€μ΄ν°λΈ 컀λ λ°μ΄λ리 (1.24 MiB)config.json: μ λ‘ νΈλ¬μ€νΈ μκ³μΉ λ° μ λ ₯ νΌμ² Robust Scaler λ©νλ°μ΄ν°
π License & Commercial Terms (λΌμ΄μ μ€ λ° μμ μ λμ μλ΄)
Non-Commercial Use Only (λΉμμ μ μ°κ΅¬Β·κ°μΈ μ΄μ© νμ ):
This model is strictly provided for non-commercial, research, and educational purposes only. Any commercial use without a separate commercial agreement is strictly prohibited.
λ³Έ λͺ¨λΈμ μμ μ°κ΅¬, κ΅μ‘ λ° κ°μΈ λΉμ리 λͺ©μ μ νν΄μλ§ λ¬΄λ£λ‘ μ 곡λ©λλ€. μ¬μ νκ° μλ μΌμ²΄μ μ리μ μ΄μ©μ μ격ν κΈμ§λ©λλ€.Commercial Licensing & Inquiries for Revenue-Generating Entities (λ§€μΆ λ°μ κΈ°μ λ° μμ©ν λμ λ¬Έμ):
If your organization generates revenue, or if you plan to incorporate this model into commercial products, services, or paid APIs, you are required to purchase a commercial license. Please contact us directly via email for enterprise licensing and integration support.
κΈ°μ /μ¬μ 체μ λ§€μΆμ΄ λ°μνκ³ μκ±°λ, λ³Έ λͺ¨λΈμ μμ© μλΉμ€Β·μ λ£ μ νΒ·μμ΅ μ°½μΆ λͺ©μ μ μμ€ν μ λμ νκ³ μ νμλ κ²½μ° λ°λμ λ³λμ μμ© λΌμ΄μ μ€λ₯Ό μ·¨λνμ μΌ ν©λλ€. λμ 쑰건 λ° κΈ°μ μ© λΌμ΄μ μ€ λ°κΈμ μλ μ΄λ©μΌλ‘ λ¬Έμν΄ μ£ΌμκΈ° λ°λλλ€.- λμ λ° μμ© λΌμ΄μ μ€ λ¬Έμ (Commercial Inquiries): admin@099.kr
- 곡μ μΉμ¬μ΄νΈ (Official Website): https://baar.uk/model/
β οΈ Notes & Disclaimer (κΈ°μ κ³ μ§ λ° μλ΄)
- Defensive Engineering & Evaluation: λ³Έ λͺ¨λΈμ μΈλΌμΈ μΉ¨μ
λ°©μ§ μμ€ν
(IPS) λ° μ λ‘ νΈλ¬μ€νΈ μν€ν
μ² μ°κ΅¬λ₯Ό μν΄ μ μλμμ΅λλ€. μ€μ μ΄μ λ€νΈμν¬μ λ°°μΉ μ, μ‘°μ§μ 보μ μ μ±
μ λ§μΆ° μ€ν νμ©μΉμ μ°¨λ¨ μκ³κ°(
config.json)μ μ¬μ μ μ‘°μ νμ¬ μ μ©νμμμ€.
πΌ Opportunities & Contact (μ±μ© μ μ λ° ννΈλμ)
This project demonstrates end-to-end competency in adversarial network traffic defense, multi-source dataset engineering, PGD adversarial training, and ultra-low latency inference packaging (.baar3, ONNX).
I am actively seeking AI Engineering / Security AI Research opportunities, recruitment offers, and technical partnerships.
- 곡μ μΉμ¬μ΄νΈ (Official Website): https://baar.uk/model/
- κΈ°μ λμ λ° λΌμ΄μ μ€ λ¬Έμ (Commercial Adoption): admin@099.kr
- μ±μ© λ° ν¬μ§μ μ μ (Recruitment & Hiring): admin@099.kr
- ν¬μ λ° κΈ°μ νλ ₯ λ¬Έμ (Investment & Partnerships): admin@099.kr
κ·Ήλ¨μ μΈ λ¦¬μμ€ νκ²½μμλ κ³ μ μΆλ‘ μ΄ κ°λ₯ν λ₯λ¬λ μν€ν μ²λ₯Ό μ€κ³νκ³ , μ€μ μ¬μ΄λ² μν λ°μ΄ν° μ΅ν© νμ΄νλΌμΈλΆν° μ λμ νλ ¨ λ° μμν λ°μ΄λ리 λ°°ν¬κΉμ§ μ κ³Όμ μ μ§μ ꡬννλ AI μμ§λμ΄μ λλ€. κΈ°μ λμ , μ±μ© μ μ λ° νμ μ κ΄μ¬ μλ κΈ°μ κ³Ό νμ λ¬Έμλ₯Ό νμν©λλ€.
- Downloads last month
- 40