ICACT20260343 Question.1
Questioner: tanghr@dsp.ac.cn 2026-02-07 오후 7:22:12 |
ICACT20260343 Answer.1
Answer by Auhor hejf@dsp.ac.cn 2026-02-07 오후 7:22:12
Chrome  Click!! |
| How does the federated learning module integrate local anomaly detection with global model updates without sharing raw data? |
In our framework, each resolution domain trains a lightweight anomaly detector locally on features extracted from its own resolution/caching behaviors, keeping raw traffic and logs on-site. During each federated round, the node sends only the updated model parameters (carrying a round identifier and sender signature) to the controller, which aggregates them with FedAvg to produce a new global model and distributes it back. The returned global model is then used for cross-domain anomaly detection and can also feed into the trust/credit adjustment and mitigation policies—without ever sharing raw data.
|
ICACT20260343 Question.2
Questioner: wangsy@dsp.ac.cn 2026-02-07 오후 7:25:07 |
ICACT20260343 Answer.2
Answer by Auhor hejf@dsp.ac.cn 2026-02-07 오후 7:25:07
Chrome Click!! |
| The Name Resolution Service (NRS) mentioned in the abstract plays what role in the ICN? And why is it targeted by attacks? |
In NRS-based ICN, the Name Resolution Service (NRS) provides the control-plane function of mapping content names/identifiers to locators/forwarding hints and distributing those resolution results to forwarding devices, which improves scalability and flexibility by decoupling resolution/control from data forwarding. It becomes a prime DDoS target because resolution state and decision points concentrate in the NRS and along the resolution–control loop, so attackers can overload it (and downstream forwarding tables) by generating bursts of resolution queries and rule updates, and by driving cache/forwarding stress such as cache pollution and forwarding exhaustion.
|
ICACT20260343 Question.3
Questioner: wangsy@dsp.ac.cn 2026-02-07 오후 7:26:51 |
ICACT20260343 Answer.3
Answer by Auhor hejf@dsp.ac.cn 2026-02-07 오후 7:26:51
Chrome Click!! |
| What is the core approach of the "trust-based defense" proposed in the paper? (Assigning scores to entities? Issuing certificates? Linking speed limits/cache policies/forwarding strategies?) |
The paper’s “trust-based defense” is a closed-loop trust control: it (1) binds each requester to a DID and uses lightweight credentials (e.g., JWTs) to cheaply authenticate requests on the fast path, (2) computes a multi-factor credit/reputation score from the requester’s resolution success and caching behavior, (3) synchronizes trust across domains via lightweight gossip, and then (4) maps the credit directly into enforcement—high-credit entities get priority and looser limits, while low-credit/malicious ones are throttled/blocked, and caching is biased toward high-credit traffic (low-credit-induced entries are not cached or evicted first). |
ICACT20260343 Question.4
Questioner: md02anas02@gmail.com 2026-02-09 오후 10:28:32 |
ICACT20260343 Answer.4
Answer by Auhor hejf@dsp.ac.cn 2026-02-09 오후 10:28:32
Chrome Click!! |
| What type of data or test environment did you use to evaluate your method? |
We evaluated the method in an ns-3 simulation testbed extended to model NRS-based ICN with explicit name resolution and SDN-style control: a topology with 20 edge nodes, 10 cache nodes, and 2 producers, where 300 benign users generate Zipf-like requests and 200 attackers generate cache-pollution/forwarding-exhaustion traffic (high-entropy names, similar-name pollution, invalid identifiers, and dynamic short-lived names). We ran each simulation for 250 s with an attack window from 40 s to 200 s, using links of 10 Mb/s and 10 ms delay, LRU caches of 1000 Data packets, and a namespace of 10,000 objects. |
ICACT20260343 Question.5
Questioner: md02anas02@gmail.com 2026-02-09 오후 10:31:06 |
ICACT20260343 Answer.5
Answer by Auhor hejf@dsp.ac.cn 2026-02-09 오후 10:31:06
Chrome Click!! |
| What is the main advantage of your proposed method compared to existing approaches? |
The main advantage is that our framework improves both security and service quality at the same time: by combining trust/credit-guided enforcement (rate limiting + caching + forwarding) with cross-domain federated anomaly learning, it achieves the highest detection accuracy (F1 about 0.94–0.96 vs. 0.82–0.92 for strong baselines) while keeping the network usable during attacks (RSR ~88–90% and CHR ~70%, whereas baselines often drop below 80% and 65%), with only lightweight coordination (e.g., small gossip overhead and FL sharing only model parameters). |
ICACT20260343 Question.6
Questioner: hhl9438@chungbuk.ac.kr 2026-02-09 오후 10:33:19 |
ICACT20260343 Answer.6
Answer by Auhor hejf@dsp.ac.cn 2026-02-09 오후 10:33:19
Chrome Click!! |
| You mentioned cold-start protection where new users start with a neutral credit score and conservative rate limits. Could this approach negatively impact legitimate new users causing high traffic, such as a flash crowd for a popular event? |
Yes—if you set the “cold-start” limits too aggressively, a legitimate flash crowd could be throttled initially; however, in our design newly enrolled entities start at a neutral credit and are only kept under conservative limits until enough observations accumulate. Credit is updated over a sliding window using recent + long-term resolution success and cache-hit behaviors, and the enforcement layer maps higher credit to looser rate limits and higher forwarding priority, so legitimate high-demand users should “earn” capacity quickly as they generate successful, cache-friendly traffic. For planned events, the same enrollment step (domain authority endorsing DIDs/credentials) can also be used to pre-provision trusted clients to reduce cold-start impact. |
