第 144 / 209 页
非代码面试题
显示 20 / 4169 道匹配题目
答题状态：未尝试未正确已正确
ID题目领域难度题型进度权限
4327CNN Depth For Longer HorizonA stride-1 CNN uses kernel size 3 and no dilation. To cover a dependency horizon of 9 steps you need 4 layers. If the required horizon rises to 41 steps, how many layers are needed, and what does that imply about the architecture pressure?机器学习中等essay未尝试面试订阅4328Small-Data Regime ShiftSuppose the task stays strongly local and translation-equivariant, but your labeled dataset shrinks by a factor of 10. Which architecture becomes more attractive, and why does the shift in data regime matter?机器学习中等essay未尝试面试订阅4329Latency Budget RelaxationA task was originally fully online, making recurrence or causal convolution preferable. If the deployment changes to offline batch scoring with the whole sequence available, which architecture family gains the most from that relaxation?机器学习中等essay未尝试面试订阅4330From Local To Global Task StructureA prediction problem used to depend on short motifs, but after a product change the label now depends on matching information from the first and last quarter of each sequence. Which architecture family should move up the ranking?机器学习中等essay未尝试面试订阅4331What To Quantify FirstBefore you choose between a CNN, RNN, and Transformer for a new sequence task, what two structural quantities should you quantify first?机器学习中等essay未尝试面试订阅4332Before Picking Transformer By DefaultA teammate wants to start with a Transformer because it won the last benchmark. What is the first counter-question you should ask?机器学习中等essay未尝试面试订阅4333Before Discarding RNNsWhy should you hesitate before ruling out RNNs entirely in a trading-system pipeline?机器学习中等essay未尝试面试订阅4334Before Using CNNWhat is the first structural property you should verify before leaning on a CNN as your main architecture?机器学习中等essay未尝试面试订阅4335Hybrid ThinkingIf you suspect the task has both strong local motifs and occasional long-range dependencies, what should be your first decomposition step before arguing about model family?机器学习中等essay未尝试面试订阅4336Why CNN Can WinWhy can a modest CNN beat a larger Transformer on a small-data task whose label depends mainly on short local patterns?机器学习中等essay未尝试面试订阅4337Why RNN Still MattersWhy might an RNN still be the practical choice for a production event-stream model even if Transformers benchmark better offline?机器学习中等essay未尝试面试订阅4338When Attention Earns Its CostWhat kind of task structure makes the quadratic cost of attention worth paying?机器学习中等essay未尝试面试订阅4339Why Architecture Mismatch HurtsWhy can architecture mismatch dominate parameter count when performance is poor?机器学习中等essay未尝试面试订阅4340Hybrid Versus PureWhen is it more sensible to consider a hybrid architecture instead of insisting on a pure CNN, pure RNN, or pure Transformer?机器学习中等essay未尝试面试订阅4341Confusion-Matrix Metrics 1At a fixed threshold, prevalence is 20%, TPR is 80%, and FPR is 10%. What precision does that imply?机器学习简单数值题未尝试面试订阅4342Confusion-Matrix Metrics 2A fraud model keeps TPR = 0.90 and FPR = 0.03 when deployed into a market where prevalence falls from 10% to 2%. What precision should you now expect at the same threshold?机器学习简单数值题未尝试面试订阅4343Confusion-Matrix Metrics 3Predicted probabilities are [0.8, 0.6, 0.3, 0.1] and labels are [1, 0, 1, 0]. What is the Brier score?机器学习简单数值题未尝试面试订阅4344Confusion-Matrix Metrics 4For expected calibration error with equal sample weighting, you have two nonempty bins. Bin A has probabilities [0.2, 0.3] and labels [0, 1]. Bin B has probabilities [0.8, 0.9] and labels [1, 1]. Using ECE = sum over bins of (bin fraction)*|avg confidence - accuracy|, what ECE do you get?机器学习简单数值题未尝试面试订阅4345Confusion-Matrix Metrics 5A model assigns an average predicted probability of 0.18 to a bucket containing 200 names. If the model is calibrated, how many positives should you expect in that bucket on average?机器学习简单数值题未尝试面试订阅4346Brier Score Snapshot 1At one threshold, prevalence is 5%, TPR is 80%, and FPR is 10%. What PR-space point (recall, precision) corresponds to that ROC-space operating point?机器学习中等数值题未尝试面试订阅