SPRINT: Semi-supervised Prototypical Representation for Few-Shot Class-Incremental Tabular Learning

Umid Suleymanov; Murat Kantarcioglu; Kevin S Chan; Michael De Lucia; Kevin Hamlen; Latifur Khan; Sharad Mehrotra; Ananthram Swami

cs.CVMar 4, 2026

SPRINT: Semi-supervised Prototypical Representation for Few-Shot Class-Incremental Tabular Learning

Authors: Umid Suleymanov, Murat Kantarcioglu, Kevin S Chan, Michael De Lucia, Kevin Hamlen, Latifur Khan, Sharad Mehrotra, Ananthram Swami,

Organizations: Department of Computer Science, Virginia Tech, Blacksburg, VA, USA · Department of Computer Science, University of Texas at Dallas, Dallas, TX, USA · DEVCOM Army Research Laboratory, Adelphi, MD, USA · Department of Computer Science, University of California, Irvine, CA, USA

Abstract

Real-world systems must continuously adapt to novel concepts from limited data without forgetting previously acquired knowledge. While Few-Shot Class-Incremental Learning (FSCIL) is established in computer vision, its application to tabular domains remains largely unexplored. Unlike images, tabular streams (e.g., logs, sensors) offer abundant unlabeled data, a scarcity of expert annotations and negligible storage costs, features ignored by existing vision-based methods that rely on restrictive buffers. We introduce SPRINT, the first FSCIL framework tailored for tabular distributions. SPRINT introduces a mixed episodic training strategy that leverages confidence-based pseudo-labeling to enrich novel class representations and exploits low storage costs to retain base class history. Extensive evaluation across six diverse benchmarks spanning cybersecurity, healthcare, and ecological domains, demonstrates SPRINT's cross-domain robustness. It achieves a state-of-the-art average accuracy of 77.37% (5-shot), outperforming the strongest incremental baseline by 4.45%.

Figures & tables

Figure 1: Schematic overview of the SPRINT framework.

Dataset	Feat.	Base Classes ( $C_{base}$ )	Incr. Classes ( $C_{novel}$ )	$M^{(0)}$ / Class
ACI-IoT-2023 ( Bastian et al., 2023 )	79	Benign, Dictionary, OS Scan, Port Scan, Ping Sweep, Vuln. Scan	ICMP Flood, Slowloris, SYN Flood, DNS Flood	2,000
CIC-IDS2017 ( Sharafaldin et al., 2018 )	77	BENIGN, PortScan, DDoS, FTP-Patator, SSH-Patator	DoS Hulk, DoS GoldenEye, DoS Slowhttptest, DoS slowloris	2,000
CIC-IoT2023 ( Neto et al., 2023 )	39	BENIGN, DNS Spoofing, DoS: HTTP, SYN, TCP, UDP, Mirai: Greeth, Greip, UDPPlain, MITM, Recon: Host, OS, Port, Vuln. Scan	DDoS: Ack, HTTP, ICMP, ICMP-FRAG, PSHACK, RSTFIN, SynonymousIP, SYN, TCP, UDP, UDP-Frag	2,000
Obesity ( Palechor and la Hoz Manotas, 2019 )	31	Insufficient Wgt, Normal Wgt, Overweight Lvl I, Overweight Lvl II	Obesity Type I, Obesity Type II, Obesity Type III	100
CovType ( Blackard, 1998 )	54	Spruce/Fir, Lodgepole Pine, Ponderosa Pine, Cottonwood/Willow	Aspen, Douglas-fir, Krummholz	2,000
MNIST ( LeCun et al., 2010 )	784	Digits: 0, 1, 2, 3, 4, 5	Digits: 6, 7, 8, 9	2,000

Table 1: Detailed configuration of dataset splits, class partitions, and base class memory budgets (

M^{(0)}

samples per base class) used in the FSCIL experiments. The Incr. Classes column lists categories introduced sequentially in sessions

1\dots S

Method	ACI-IoT-2023		Obesity		CICIDS2017		CIC-IoT-2023		CovType		MNIST		Average
Method	PD $\downarrow$	Acc $\uparrow$	PD $\downarrow$	Acc $\uparrow$	PD $\downarrow$	Acc $\uparrow$	PD $\downarrow$	Acc $\uparrow$	PD $\downarrow$	Acc $\uparrow$	PD $\downarrow$	Acc $\uparrow$	Avg-PD $\downarrow$	Avg-Acc $\uparrow$
CNN ProtoNet	9.49	76.22	5.04	55.53	23.51	69.91	0.99	53.56	9.72	45.87	8.89	81.10	9.61	63.70
FACT	8.06	69.93	10.71	14.29	15.17	18.42	-2.46	53.68	13.47	16.84	25.00	40.65	11.66	35.63
MAML	33.92	47.27	23.95	37.28	43.96	42.67	18.22	35.39	32.62	38.28	45.59	48.42	33.04	41.55
Neuron Expansion	12.33	85.79	22.60	59.45	20.92	74.01	30.71	60.39	30.33	50.08	27.13	70.90	24.00	66.77
ProtoNet	12.03	84.16	10.71	71.67	27.05	69.21	2.02	59.85	11.44	54.91	23.87	70.97	14.52	68.46

Table 2: Summary of results on 6 datasets (5-shot setting). PD: Performance Dropping rate (

A_{0}-A_{last}

), Acc: Last session accuracy (

A_{last}

). Avg-PD and Avg-Acc represent the average values across all datasets. Best results are in bold , second best in italics .

Panel of Figure 2 — Figure 2: Session-wise accuracy comparison on the a) CICIDS2017 and b) MNIST datasets. The annotation highlights the performance margin of SPRINT over the second-best method in the final incremental session.

Figure 3: Distribution of Top-1 Accuracy (Final Session). Kernel Density Estimate (KDE) and histogram comparison of SPRINT vs. iCaRL over 30 independent runs on the ACI-IoT-2023 dataset. SPRINT exhibits a sharp, high-confidence peak around 93.6%, demonstrating superior stability and minimal variance compared to the broader distribution of iCaRL.

Panel of Figure 4 — Figure 4: Design Analysis and Ablation Studies. (a) Impact of distance metrics on classification accuracy. (b) Evaluation in non-incremental setting. (c) Sensitivity of the loss balancing term $\beta$ . (d) Influence of the number of pseudo-samples on Top-1 accuracy.

Panel of Figure 5 — Figure 5: Sensitivity and Component Analysis. (a) Influence of the memory budget $M^{(0)}$ on final retention. (b) model accuracy vs shot number. (c) Breakdown of pseudo-label accuracy, demonstrating the effectiveness of the semi-supervised component.

Method	5-shot						10-shot
Method	0	1	2	3	4	Imp.	0	1	2	3	4	Imp.
CNN ProtoNet	85.70	82.47	80.96	81.94	76.22	+17.42	88.25	85.33	84.49	85.14	79.91	+15.35
FACT	77.99	76.23	74.73	75.85	69.93	+23.71	84.45	84.60	83.47	84.76	78.39	+16.88
MAML	81.18	66.98	60.06	54.87	47.27	+46.37	83.43	68.98	61.29	55.50	47.93	+47.34
Neuron Expansion	98.11	88.58	88.38	88.05	85.79	+7.85	98.16	92.28	90.71	91.06	89.98	+5.29
ProtoNet	96.19	89.15	86.33	87.06	84.16	+9.47	97.23	90.86	87.57	88.08	85.25	+10.02

Table 3: Comparison of different models on ACI-IoT-2023 dataset (5-shot and 10-shot). Values indicate mean accuracy (%). Best results per session are in bold , second-best in italics . The ‘Imp.’ column shows the accuracy improvement of SPRINT over each method.

Method	1	2	3	4
Baseline	89.15	86.33	87.06	84.16
+ $\mathcal{L}_{proto}$	88.84	85.77	86.09	82.64
+ $\mathcal{L}_{semi-clust.}$	91.01	89.18	88.84	85.13
SPRINT	96.17	96.29	95.91	93.63

Table 4: Ablation study accuracy (%) on ACI-IoT-2023 (5-shot).

Appendix figures & tables12 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 6: Incremental Training Latency Comparison. Training runtime (in seconds) averaged across all incremental sessions and 10 independent runs. Numbers above bars indicate the speedup factor (

T_{\text{iCaRL}}/T_{\text{SPRINT}}

). SPRINT consistently achieves an

\approx 18\times

reduction in training latency due to its sparse episodic sampling strategy, which decouples computational cost from the total memory buffer size.

Method	100 Epochs	200 Epochs	300 Epochs	Speedup
iCaRL	$57.09\pm 1.87$	$111.78\pm 0.98$	$170.21\pm 0.48$	$1.0\times$
SPRINT	$\mathbf{3.12}\pm 0.15$	$\mathbf{5.97}\pm 0.14$	$\mathbf{8.87}\pm 0.17$	$\mathbf{18.3\times}$

Appendix

Table 5: Training time profiling (seconds). Results averaged over 10 independent runs (

\pm

std dev).

Parameter	Value	Description
Evaluation Protocol
Independent Runs	30	Number of distinct random seeds used for reporting means
Test Episodes	500	Number of episodes evaluated per run
Support Set Size	$K$	Matches the $K$ -shot setting ( Snell et al., 2017 )
Query Points ( $N_{Q}$ )	15	Number of query samples per class per episode
Training & Optimization

Appendix

Table 6: Summary of implementation hyperparameters and evaluation protocols.

Figure 7: Impact of Training Duration on Neuron Expansion. Performance sensitivity (Top-1 Accuracy and F1 Score) of the Neuron Expansion model across varying epoch counts on the ACI-IoT-2023 dataset. The model achieves peak stability around 30 epochs, after which performance plateaus or slightly degrades, indicating limited benefit from prolonged training.

Method	ACI-IoT-2023		Obesity		CICIDS2017		CIC-IoT-2023		CovType		MNIST		Average
Method	PD	Acc	PD	Acc	PD	Acc	PD	Acc	PD	Acc	PD	Acc	Avg-PD	Avg-Acc
CNN ProtoNet	8.33	79.91	5.53	60.32	21.02	73.34	0.68	56.17	9.39	49.21	7.11	86.09	8.68	67.51
FACT	6.06	78.39	11.15	15.19	13.29	18.71	-3.04	53.83	12.70	16.40	22.63	40.59	10.47	37.19
MAML	35.50	47.93	26.13	38.62	43.53	43.31	19.44	34.44	32.86	39.19	44.17	49.89	33.61	42.23
Neuron Expansion	8.18	89.98	14.40	67.74	17.54	77.22	28.77	62.15	28.60	51.94	19.40	78.68	19.48	71.29
ProtoNet	11.98	85.25	10.35	72.71	26.17	70.55	2.45	61.40	10.94	57.24	22.27	73.23	14.03	70.06

Appendix

Table 7: Summary of results on 6 datasets (10-shot setting). PD: Performance Dropping rate (

A_{0}-A_{last}

), Acc: Last session accuracy (

A_{last}

). Avg-PD and Avg-Acc represent the average values across all datasets. Best results are in bold , second best in italics .

Figure 8: Distribution of Top-1 Accuracy (Final Session,

t=4

). Kernel Density Estimate (KDE) and histogram comparison of SPRINT vs. iCaRL over 30 independent runs on the CovType dataset. SPRINT shows a substantial performance shift (

\Delta\approx 4.9\%

) with a large effect size (

d=4.60

), exhibiting a distribution that is completely separated from the baseline.

Model	Compared To	Mean $\Delta$ (%)	Cohen’s $d$	$p$ (Bonf.)	Sig.
SPRINT	MAML	46.366	23.14	$<$ 0.001	Yes
SPRINT	FACT	23.707	1.88	$<$ 0.001	Yes
SPRINT	CNN ProtoNet	17.417	12.40	$<$ 0.001	Yes
SPRINT	ProtoNet	9.471	8.94	$<$ 0.001	Yes
SPRINT	Semi-Super-Proto	8.502	5.18	$<$ 0.001	Yes
SPRINT	Neur. Exp.	7.846	3.32	$<$ 0.001	Yes

Appendix

Table 8: Bonferroni-corrected one-sided p-values and effect sizes (Cohen’s

d

) for SPRINT vs. competing models on ACI-IoT-2023 (5-shot, Final Session). Results are based on 30 independent runs. Bold indicates statistical significance (

p<0.05

Dataset	Mean $\Delta$ (%)	Cohen’s $d$	$p$ (Bonf.)	Sig.
ACI-IoT-2023	4.455	2.98	$<$ 0.001	Yes
CIC-IDS2017	2.571	1.11	$<$ 0.001	Yes
CIC-IoT-2023	-5.421	-4.02	1.00	No
CovType	4.897	4.60	$<$ 0.001	Yes
Obesity	3.959	1.97	$<$ 0.001	Yes
MNIST	13.852	6.96	$<$ 0.001	Yes

Appendix

Table 9: Summary of statistical significance tests comparing SPRINT vs. iCaRL (the strongest baseline) across all datasets. Positive

\Delta

indicates SPRINT outperforms iCaRL.

Model	Compared To	Mean $\Delta$ (%)	Cohen’s d	p (Bonf.)	Signif. (Bonf.)
SPRINT	MAML	40.129	16.57	$<$ 0.001	Yes
SPRINT	FACT	64.381	17.88	$<$ 0.001	Yes
SPRINT	Semi-Super-ProtoNet	15.332	6.70	$<$ 0.001	Yes
SPRINT	ProtoNet	13.593	5.59	$<$ 0.001	Yes
SPRINT	CNN ProtoNet	12.892	5.72	$<$ 0.001	Yes
SPRINT	Neuron Expansion	8.790	1.87	$<$ 0.001	Yes

Appendix

Table 10: Bonferroni-corrected one-sided p-values and effect sizes (Cohen’s d) vs SPRINT on 5-shot last-session accuracies (CICIDS2017). All tests conducted at a 95% confidence level.

Figure 9: Session-wise Performance Distribution and Stability. Comparison of Top-1 accuracy distributions across 30 independent runs on the CICIDS2017 dataset for (left) 5-shot and (right) 10-shot settings. The boxplots visualize the variance and median performance at each incremental session. SPRINT (purple) demonstrates superior stability with minimal variance and negligible performance decay compared to baselines like iCaRL and MAML, which exhibit significant inter-run volatility and catastrophic forgetting.

Figure 10: Distribution of Incremental Performance (5-Shot). Comparative violin plots of session-wise accuracy for STUNT ( Nam et al., 2023 ) and SPRINT across five datasets. The distributions reveal a distinct performance gap: SPRINT consistently operates in a high-accuracy regime (top quartiles), whereas STUNT remains stagnated at lower accuracy levels.

Dataset	STUNT		SPRINT
Dataset	PD (%) $\downarrow$	Acc (%) $\uparrow$	PD (%) $\downarrow$	Acc (%) $\uparrow$
ACI-IoT-2023	-0.23	72.26	2.54	93.63
CIC-IoT-2023	-3.42	54.95	-3.94	65.81
CICIDS2017	10.99	57.58	13.45	82.80
CovType	12.91	45.65	10.62	58.36
MNIST	12.11	57.56	5.18	84.85

Appendix

Table 11: Comparison of self-supervised baseline STUNT versus SPRINT on 5-shot incremental settings. While STUNT exhibits low PD in some cases due to poor initial base learning, SPRINT consistently achieves superior final accuracy (

A_{last}

) while maintaining low performance degradation.

Class-Incremental Learning

Class-incremental learning studies how models can add new categories over time while retaining the ability to recognize previously learned classes.

See all 162 papers

Decoupled and Distilled: Task-Adaptive LoRA-Teachers with Ensemble Knowledge Transfer for Few-Shot Class-Incremental Learning
Sep 30, 2026Hongwei Zhao, Rui Liu, Yansong Liu +2Class-Incremental Learning Low-Rank Adaptation
School of Computer Science and Engineering, Beihang University · School of Computer Science, Beijing University of Posts and Telecommunications
Selective Backpropagation for Efficient Few-Shot Class-Incremental Learning
Oct 2, 2026Eeham Khan, Abdulmoumen Al-Atrash, Ali AyubClass-Incremental Learning Neural Network Optimization
PaInt Lab, Concordia University Montreal, QC, Canada
Learning to Curate What You Generate for Generalizable Few-Shot Class-Incremental Learning
Oct 4, 2026Junhui Yin, Yuchen Yang, Yilin Yin +6Class-Incremental Learning Data Selection
Beijing University of Posts and Telecommunications, Beijing, China · Institute of Automation, Chinese Academy of Sciences, Beijing, China · Singapore Management University, Singapore, Singapore

Sep 30, 2026J/K move · Enter open · S save

Decoupled and Distilled: Task-Adaptive LoRA-Teachers with Ensemble Knowledge Transfer for Few-Shot Class-Incremental Learning

Hongwei Zhao, Rui Liu, Yansong Liu, Zhiyuan Zou+1

School of Computer Science and Engineering, Beihang University · School of Computer Science, Beijing University of Posts and Telecommunications

Class-Incremental Learning Low-Rank Adaptation Multi-Teacher Knowledge Distillation Few-Shot Image Classification Knowledge Distillation

Open paper AskPDF arXiv

Few-Shot Class-Incremental Learning (FSCIL) addresses the challenge of learning new classes from very limited samples while retaining knowledge of previously learned ones. Although parameter-efficient fine-tuning methods with pre-trained models show promise for class-incremental learning, strict gradient-based constraints can be unreliable under severe data scarcity, while multi-expert approaches can impose substantial inference-time costs. We propose TALON (Task-Adaptive LoRA-Teachers with Ensemble Knowledge Transfer), an inference-efficient FSCIL framework. TALON dynamically allocates an independent LoRA-Teacher to each incremental task for task-specific representation learning, then distills multiple frozen teachers into a unified LoRA-Student through Ensemble Knowledge Transfer, eliminating runtime module selection or generation. A semantic-guided distillation strategy weights teacher contributions by feature-space similarity to mitigate catastrophic forgetting and overfitting. Across three class-order runs, TALON achieves comparable or better mean average accuracy across four FSCIL benchmarks, obtaining 86.68 +/- 1.22% on CUB200, 90.39 +/- 0.27% on CIFAR100, 78.38 +/- 0.94% on ImageNet-R, and 96.34 +/- 0.33% on miniImageNet. TALON uses up to 33x fewer deployment parameters and reduces average inference time per task to 26.7 s, a 41.70% reduction relative to ASP.

Abstract

Class-Incremental Learning

Explore similar work

Decoupled and Distilled: Task-Adaptive LoRA-Teachers with Ensemble Knowledge Transfer for Few-Shot Class-Incremental Learning

Selective Backpropagation for Efficient Few-Shot Class-Incremental Learning

Learning to Curate What You Generate for Generalizable Few-Shot Class-Incremental Learning