AI models, and time-series models in particular are typically ranked by accuracy and deployed under constraints absent from that ranking. In this paper, we study a real setting where these constraints are explicit and measurable: handover pre- diction in cellular networks, where inference runs on the User Equipment once per 40 ms measurement period and must complete before the network acts. The timing bound is hard: a late prediction is superseded by the measurement it was predicting, while energy is a cumulative budget spent for the life of the device. Neither appears in the objective these models are selected on. We evaluate eight architectures under a protocol that screens for deadline feasibility at the tail rather than the mean, then ranks the feasible models on measured inference energy, count- ing the feature computation that precedes the forward pass and is usually omitted. Gradient-boosted trees on handcrafted temporal features are the most accurate model (F1=0.974) and among the cheapest to run, while the neural architecture an accuracy-only benchmark would rank first (CNN-1D, F1=0.914) costs over 3 times more energy per inference under a realistic duty-cycled deployment pattern. We give three conditions under which this outcome should be expected, so the finding applies beyond telecommunications. Accuracy-only benchmarking does not merely under-report deployment cost. It can rank first a model that cannot be deployed.
Figures & tables
Latency (ms)
Energy ( μ J)
Model
Params
F1
AUC
p50
p99
Total p50
Looped
Duty
L/D
LightGBM
16,317
0.974 ± 0.002
0.998
0.250
0.381
0.263
1,473
4,947
0.30
CNN-1D
81,217
0.914 ± 0.012
0.985
0.947
1.095
0.892
15,316
16,181
0.95
Efficient-HO-v2
16,753
0.899 ± 0.007
0.984
1.252
1.388
1.372
19,940
20,285
0.98
MiniRocket
9,996
0.872 ± 0.011
0.972
1.321
1.472
1.329
39,596
44,611
0.89
Transformer
70,465
0.870 ± 0.019
0.973
0.770
0.899
0.833
10,643
14,635
0.73
Table 1: Main benchmark on the 100 s dataset (five seeds, mean ± std). Latency and energy are full-pipeline (preprocess + forward + postprocess). Energy is per-inference; Looped keeps the core busy (throughput benchmark), Duty runs inferences in bursts with idle pauses as deployed, and the ratio quantifies how much looped measurement understates deployment energy. Best neural in italics , overall best in bold .
Figure 1: Energy-accuracy trade-off. LightGBM dominates all neural models in both F1 and inference energy.
Appendix figures & tables3 assets
Supplementary material from the paper’s appendix.
Appendix
Model
Params
Family
Configuration
LSTM
57,665
recurrent
2 layers, hidden 64, dropout 0.2
Transformer
70,465
attention
dmodel=64 , 2 layers, 4 heads
CNN-1D
81,217
convolutional
3 residual blocks, 64 channels, kernel 3
LightGBM
16,317
trees + features
500 rounds, 63 leaves, 247 features
MiniRocket
9,996 feat.
random features
10k kernels, ridge
catch22+ridge
462 feat.
feature + linear
22 features × 21 cells, ridge
Appendix
Table 4: Evaluated models. Parameter counts for the tree models are total leaves across all boosting rounds; MiniRocket and catch22 use a ridge classifier, so their count is the feature dimension.
Model
Params
Train Time (s)
Energy (kWh)
CO 2 (g)
Best Val Acc
LightGBM
16,317
00 4.2 ± 0.2
1.3×10−4
0.005
0.981
LSTM
57,665
122.1 ± 5.5
3.5×10−3
0.141
0.885
Efficient-HO
0 7,618
134.4 ± 5.9
3.8×10−3
0.156
0.808
Efficient-HO-v2
16,753
146.8 ± 13.9
4.8×10−3
0.196
0.942
CNN-1D
81,217
170.6 ± 10.2
4.9×10−3
0.198
0.945
Transformer
70,465
208.9 ± 5.1
5.9×10−3
0.242
0.929
Appendix
Table 5: Training cost on server CPU, measured with CodeCarbon (five seeds, mean ± std). Same platform as the inference results.
Algorithm
Total HOs
Success
Failures
Ping-Pong
Early Exec
TTT Saved (ms)
Algo 0 (baseline)
489
489
0
5
—
—
Algo 4 (early exec)
238
238
0
2
223
542
Appendix
Table 6: HO simulation: baseline A3 (Algo 0) vs. LSTM early execution (Algo 4) at best configuration, TTT=640 ms.
Enterprise wireless access points (APs) are promising platforms for predictive machine learning (ML), but their primary responsibility remains providing wireless connectivity and network services. Predictive inference must therefore share an AP's CPU and memory with packet processing, Wi-Fi and IoT radio operations, and client management. This resource contention creates two risks: a model that performs well on proxy hardware may be too slow on the target AP, while a model that fits in isolation may still degrade network services under load. We define network-aware deployability using two gates: qualification of the model and its execution path on the target AP, followed by validation of its execution profile under packet-service and forecasting constraints. Our benchmarks show that edge testbeds do not reliably capture target behavior. Across matched artifacts and serving settings, five model implementations run 6.1--19.1× slower on an AP than on a Raspberry Pi5, while peak memory usage differs by up to 22%. Moreover, two forecasting foundation models of similar size differ in AP latency by 19×. When serving a smaller model across 13 parallel streams at a 30s cadence under network saturation, default execution increases p99 round-trip time (RTT) by 76% and reduces throughput by 7.06%. Understanding these trade-offs is essential for live deployment if we aim to use APs for both networking and ML workloads.
Accurate pre-deployment estimation of CNN inference cost--energy, latency, and peak memory--is increasingly critical as models are deployed on resource-constrained GPU platforms. Existing approaches rely on FLOPs, latency measurements, or single-device profiling as energy proxies, overlooking the non-linear interactions between architectural design and hardware load. We present a workload characterization study of 13 419 CNN configurations on two GPU platforms (RTX 5090 and RTX 3080) under GPU telemetry, revealing that energy, latency, and memory exhibit fundamentally distinct scaling behaviors: energy and latency diverge by 3x under high computational demand, and cross-GPU transferability differs by target--energy and latency require platform-specific models while memory transfers well across the two tested platforms. Building on these characterization findings, we develop CARB, a cascade-blended ensemble that jointly predicts all three targets with R2 ~0.99, and a two-stage deployment screening workflow that eliminates over 90% of candidates in seconds, reducing large design spaces to a Pareto-prioritized shortlist validated against real hardware.
Time-series foundation models (TSFMs) provide forecasts for operational decisions, but accuracy alone does not determine their value. Evaluating agents that use these models requires measuring decision quality and forecast cost. FWBench evaluates this capability on 1,251 electricity and cycle-hire cases using fixed forecast tools and simulated capacity contracts. Agents select models, histories and horizons, then submit capacities to minimize a stated loss-cost objective. We evaluated two hosted and eight local configurations, including small language models, and tested local models with and without TSFMs. GPT-6 Astra bought inexpensive short-horizon forecasts selectively, using 2.5% of the budget, and outperformed fixed policies when the saved decisions were scored with three loss-cost weightings. FWBench enables reproducible evaluation of how language models select and use time-series forecasts to make decisions under cost constraints.